docs / huggingface_WebShop.txt
danidarko's picture
Upload 59 files
b1d4de0 verified
# File: WebShop-master/baseline_models/agent.py
import os
import random
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import AutoTokenizer
from collections import defaultdict, namedtuple
from models.bert import BertConfigForWebshop, BertModelForWebshop
from models.rnn import RCDQN
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
State = namedtuple('State', ('obs', 'goal', 'click', 'estimate', 'obs_str', 'goal_str', 'image_feat'))
TransitionPG = namedtuple('TransitionPG', ('state', 'act', 'reward', 'value', 'valid_acts', 'done'))
def discount_reward(transitions, last_values, gamma):
(returns, advantages) = ([], [])
R = last_values.detach()
for t in reversed(range(len(transitions))):
(_, _, rewards, values, _, dones) = transitions[t]
R = torch.FloatTensor(rewards).to(device) + gamma * R * (1 - torch.FloatTensor(dones).to(device))
baseline = values
adv = R - baseline
returns.append(R)
advantages.append(adv)
return (returns[::-1], advantages[::-1])
class Agent:
def __init__(self, args):
self.tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased', truncation_side='left', max_length=512)
self.tokenizer.add_tokens(['[button], [button_], [clicked button], [clicked button_]'], special_tokens=True)
vocab_size = len(self.tokenizer)
embedding_dim = args.embedding_dim
if args.network == 'rnn':
self.network = RCDQN(vocab_size, embedding_dim, args.hidden_dim, args.arch_encoder, args.grad_encoder, None, args.gru_embed, args.get_image, args.bert_path)
self.network.rl_forward = self.network.forward
elif args.network == 'bert':
config = BertConfigForWebshop(image=args.get_image, pretrained_bert=args.bert_path != 'scratch')
self.network = BertModelForWebshop(config)
if args.bert_path != '' and args.bert_path != 'scratch':
self.network.load_state_dict(torch.load(args.bert_path, map_location=torch.device('cpu')), strict=False)
else:
raise ValueError('Unknown network: {}'.format(args.network))
self.network = self.network.to(device)
self.save_path = args.output_dir
self.clip = args.clip
self.w = {'loss_pg': args.w_pg, 'loss_td': args.w_td, 'loss_il': args.w_il, 'loss_en': args.w_en}
self.optimizer = torch.optim.Adam(self.network.parameters(), lr=args.learning_rate)
self.gamma = args.gamma
def build_state(self, ob, info):
obs_ids = self.encode(ob)
goal_ids = self.encode(info['goal'])
click = info['valid'][0].startswith('click[')
estimate = info['estimate_score']
obs_str = ob.replace('\n', '[SEP]')
goal_str = info['goal']
image_feat = info.get('image_feat')
return State(obs_ids, goal_ids, click, estimate, obs_str, goal_str, image_feat)
def encode(self, observation, max_length=512):
observation = observation.lower().replace('"', '').replace("'", '').strip()
observation = observation.replace('[sep]', '[SEP]')
token_ids = self.tokenizer.encode(observation, truncation=True, max_length=max_length)
return token_ids
def decode(self, act):
act = self.tokenizer.decode(act, skip_special_tokens=True)
act = act.replace(' [ ', '[').replace(' ]', ']')
return act
def encode_valids(self, valids, max_length=64):
return [[self.encode(act, max_length=max_length) for act in valid] for valid in valids]
def act(self, states, valid_acts, method, state_strs=None, eps=0.1):
act_ids = self.encode_valids(valid_acts)
(act_values, act_sizes, values) = self.network.rl_forward(states, act_ids, value=True, act=True)
act_values = act_values.split(act_sizes)
if method == 'softmax':
act_probs = [F.softmax(vals, dim=0) for vals in act_values]
act_idxs = [torch.multinomial(probs, num_samples=1).item() for probs in act_probs]
elif method == 'greedy':
act_idxs = [vals.argmax(dim=0).item() for vals in act_values]
elif method == 'eps':
act_idxs = [vals.argmax(dim=0).item() if random.random() > eps else random.randint(0, len(vals) - 1) for vals in act_values]
acts = [acts[idx] for (acts, idx) in zip(act_ids, act_idxs)]
(act_strs, act_ids) = ([], [])
for (act, idx, valids) in zip(acts, act_idxs, valid_acts):
if torch.is_tensor(act):
act = act.tolist()
if 102 in act:
act = act[:act.index(102) + 1]
act_ids.append(act)
if idx is None:
act_str = self.decode(act)
else:
act_str = valids[idx]
act_strs.append(act_str)
return (act_strs, act_ids, values)
def update(self, transitions, last_values, step=None, rewards_invdy=None):
(returns, advs) = discount_reward(transitions, last_values, self.gamma)
stats_global = defaultdict(float)
for (transition, adv) in zip(transitions, advs):
stats = {}
(log_valid, valid_sizes) = self.network.rl_forward(transition.state, transition.valid_acts)
act_values = log_valid.split(valid_sizes)
log_a = torch.stack([values[acts.index(act)] for (values, acts, act) in zip(act_values, transition.valid_acts, transition.act)])
stats['loss_pg'] = -(log_a * adv.detach()).mean()
stats['loss_td'] = adv.pow(2).mean()
stats['loss_il'] = -log_valid.mean()
stats['loss_en'] = (log_valid * log_valid.exp()).mean()
for k in stats:
stats[k] = self.w[k] * stats[k] / len(transitions)
stats['loss'] = sum((stats[k] for k in stats))
stats['returns'] = torch.stack(returns).mean() / len(transitions)
stats['advs'] = torch.stack(advs).mean() / len(transitions)
stats['loss'].backward()
stats['gradnorm_unclipped'] = sum((p.grad.norm(2).item() for p in self.network.parameters() if p.grad is not None))
nn.utils.clip_grad_norm_(self.network.parameters(), self.clip)
stats['gradnorm_clipped'] = sum((p.grad.norm(2).item() for p in self.network.parameters() if p.grad is not None))
for (k, v) in stats.items():
stats_global[k] += v.item() if torch.is_tensor(v) else v
del stats
self.optimizer.step()
self.optimizer.zero_grad()
return stats_global
def load(self):
try:
self.network = torch.load(os.path.join(self.save_path, 'model.pt'))
except Exception as e:
print('Error saving model.', e)
def save(self):
try:
torch.save(self.network, os.path.join(self.save_path, 'model.pt'))
except Exception as e:
print('Error saving model.', e)
# File: WebShop-master/baseline_models/env.py
import sys
import json
import random
from os.path import join, dirname, abspath
from collections import defaultdict
MODEL_PATH = dirname(abspath(__file__))
SITE_PATH = join(MODEL_PATH, '../')
sys.path.insert(0, SITE_PATH)
from web_agent_site.envs import WebAgentTextEnv
from web_agent_site.utils import *
from web_agent_site.engine.goal import get_reward
class WebEnv:
def __init__(self, args, split, server=None, id=None):
self.env = WebAgentTextEnv(observation_mode=args.state_format, server=server, filter_goals=None, limit_goals=-1, num_products=args.num, human_goals=args.human_goals, get_image=args.get_image, num_prev_obs=args.num_prev_obs, num_prev_actions=args.num_prev_actions, session_prefix=id)
if args.num is None:
if split == 'test':
self.goal_idxs = range(500)
elif split == 'eval':
self.goal_idxs = range(500, 1500)
elif split == 'train':
self.goal_idxs = range(1500, len(self.env.server.goals))
else:
self.goal_idxs = range(len(self.env.server.goals))
print(self.goal_idxs)
self.steps = 0
self.step_limit = args.step_limit
self.stats = defaultdict(int)
self.session = None
self.click_item_name = args.click_item_name
self.asin2name = {k.lower(): v['Title'].lower() for (k, v) in self.env.server.product_item_dict.items()}
self.name2asin = {v: k for (k, v) in self.asin2name.items()}
self.attributes_fail = defaultdict(int)
self.attributes_success = defaultdict(int)
self.items_clicked = defaultdict(int)
self.harsh_reward = args.harsh_reward
self.go_to_item = args.go_to_item
self.go_to_search = args.go_to_search
self.ban_buy = args.ban_buy
self.prev_ob = self.cur_ob = None
self.get_image = args.get_image
self.item_rank = -1
self.reduce_click = 1
if args.extra_search_path != '':
self.extra_search = json.load(open(args.extra_search_path))
self.extra_search = {k.strip('.'): v for (k, v) in self.extra_search.items()}
else:
self.extra_search = None
def get_search_texts(self, atts, query, inst):
if self.extra_search is not None:
if ', and price lower than' in inst:
idx = inst.find(', and price lower than')
inst_ = inst[:idx]
else:
inst_ = inst
texts = self.extra_search.get(inst_, []) + [inst.lower()]
else:
texts = [query] + [f'{att} {query}' for att in atts] + [inst.lower()]
return texts
def get_valid_actions(self):
valid_info = self.env.get_available_actions()
if valid_info['has_search_bar']:
atts = self.session['goal']['attributes']
query = self.session['goal']['query']
inst = self.session['goal']['instruction_text']
texts = self.get_search_texts(atts, query, inst)
valids = [f'search[{text}]' for text in texts]
else:
valids = []
for text in valid_info['clickables']:
if text == 'buy now' and self.ban_buy:
cur_options = len(self.session['options'])
all_options = len(self.env.server.product_item_dict[self.session['asin']]['customization_options'])
if cur_options != all_options:
continue
if text != 'search':
if self.click_item_name and text in self.asin2name:
text = 'item - ' + self.asin2name[text]
valids.append(f'click[{text}]')
if self.reduce_click and len(valids) > 20:
valids = valids[:6] + random.sample(valids[6:], 10)
if len(valids) == 0:
valids = ['finish']
return valids
def score(self):
valid_acts = self.get_valid_actions()
if 'click[description]' not in valid_acts:
return 0.0
product = self.env.server.product_item_dict[self.session['asin']]
goal = self.session['goal']
price = self.env.server.product_prices.get(self.session['asin'])
options = self.session['options']
return get_reward(product, goal, price, options)
def estimate_score(self, atts, opts, verify=False):
valid_acts = self.get_valid_actions()
assert 'click[description]' in valid_acts
desc = self.step('click[description]')[0].lower()
self.step('click[< prev]')
feat = self.step('click[features]')[0].lower()
ob = self.step('click[< prev]')[0].lower()
n_att = 0
for att in atts:
if att in desc or att in feat or att in ob:
n_att += 1
r_att = n_att / len(atts)
n_opt = 0
for opt in opts:
for act in valid_acts:
if opt in act:
n_opt += 1
break
r_opt = n_opt / len(opts)
r = (n_att + n_opt + 1) / (len(atts) + len(opts) + 1)
return (r, r_att, r_opt)
def step(self, action):
if self.click_item_name and action.startswith('click[item - ') and (action[13:-1] in self.name2asin):
valid_items = [_ for _ in self.get_valid_actions() if _.startswith('click[item - ')]
if action in valid_items:
self.item_rank = valid_items.index(action) + 1
else:
self.item_rank = -1
action = f'click[{self.name2asin[action[13:-1]]}]'
(ob, reward, done, info) = self.env.step(action)
if action.startswith('click[') and action[6:-1] in self.asin2name:
self.items_clicked[action[6:-1]] += 1
desc = self.env.step('click[description]')[0].lower()
self.env.step('click[< prev]')
feat = self.env.step('click[features]')[0].lower()
self.env.step('click[< prev]')
else:
desc = feat = ''
r_visit = 0.0
(self.cur_ob, self.prev_ob) = (ob, self.cur_ob)
if info is None:
info = {}
self.steps += 1
if self.step_limit and self.steps >= self.step_limit:
done = True
if done:
info['verbose'] = self.session.get('verbose_info', {'r_att': 0.0, 'r_option': 0.0, 'r_price': 0.0, 'r_type': 0.0, 'w_att': 0.0, 'w_option': 0.0, 'w_price': 0.0})
verbose = info['verbose']
verbose['r_harsh'] = reward == 1
verbose['r_exact'] = reward == 1 and self.session['goal']['asin'] == self.session['asin']
verbose['r_norm'] = reward / self.steps
verbose['r_visit'] = r_visit
verbose['rank_item'] = self.item_rank
if self.harsh_reward:
reward = verbose['r_harsh']
for (k, v) in self.session['actions'].items():
self.stats[f'action_{k}'] += v
cat = self.session['goal']['category']
self.stats[f'cat_{cat}'] += 1
for att in self.session['goal']['attributes']:
if att in info['verbose'].get('purchased_attrs', []):
self.attributes_success[att] += 1
else:
self.attributes_fail[att] += 1
info.update({'valid': self.get_valid_actions(), 'goal': self.env.instruction_text, 'score': reward * 10, 'estimate_score': self.score(), 'prev_ob': self.prev_ob, 'desc': desc, 'feat': feat})
if self.get_image:
image_feat = self.env.get_image()
info['image_feat'] = image_feat
return (ob, (reward + r_visit) * 10, done, info)
def reset(self, idx=None):
if idx is None:
idx = random.sample(self.goal_idxs, k=1)[0]
(ob, info) = self.env.reset(idx)
self.session = self.env.server.user_sessions[self.env.session]
if info is None:
info = {}
(self.cur_ob, self.prev_ob) = (ob, None)
info.update({'valid': self.get_valid_actions(), 'goal': self.env.instruction_text, 'score': 0, 'estimate_score': self.score(), 'prev_ob': self.prev_ob, 'desc': '', 'feat': ''})
self.steps = 0
if self.go_to_search or self.go_to_item:
name = self.session['goal']['name'].lower()
(ob, _, _, info) = self.step(f'search[{name}]')
self.stats['action_go_to_search'] += 1
if self.go_to_item:
asin = self.session['goal']['asin'].lower()
if asin in self.env.get_available_actions()['clickables']:
(ob, _, _, info) = self.step(f'click[{asin}]')
self.stats['action_go_to_item'] += 1
self.item_rank = -1
return (ob, info)
def close(self):
self.env.close()
# File: WebShop-master/baseline_models/generate_search.py
import json
import time
import torch
from tqdm import tqdm
from transformers import BartForConditionalGeneration
from train_search import get_data, get_dataset, tokenizer
if __name__ == '__main__':
model = BartForConditionalGeneration.from_pretrained('./ckpts/web_search/checkpoint-800')
model.eval()
model = model.to('cuda')
dataset = get_dataset('web_search')
dataloader = torch.utils.data.DataLoader(dataset['all'], batch_size=32)
(_, all_goals) = get_data('all')
all_dec = []
for batch in tqdm(dataloader):
output = model.generate(input_ids=batch['input_ids'].to('cuda'), attention_mask=batch['attention_mask'].to('cuda'), num_beams=10, num_return_sequences=10, max_length=512, early_stopping=True)
dec = tokenizer.batch_decode(output, skip_special_tokens=True, clean_up_tokenization_spaces=False)
assert len(dec) % 10 == 0
for i in range(len(dec) // 10):
all_dec.append(dec[i * 10:(i + 1) * 10])
assert len(all_goals) == len(all_dec)
d = {goal: dec for (goal, dec) in zip(all_goals, all_dec)}
with open('./data/goal_query_predict.json', 'w') as f:
json.dump(d, f)
# File: WebShop-master/baseline_models/logger.py
import os
import sys
import shutil
import os.path as osp
import json
import time
import datetime
import tempfile
from collections import defaultdict
import wandb
DEBUG = 10
INFO = 20
WARN = 30
ERROR = 40
DISABLED = 50
class KVWriter(object):
def writekvs(self, kvs):
raise NotImplementedError
class SeqWriter(object):
def writeseq(self, seq):
raise NotImplementedError
class HumanOutputFormat(KVWriter, SeqWriter):
def __init__(self, filename_or_file):
if isinstance(filename_or_file, str):
self.file = open(filename_or_file, 'wt')
self.own_file = True
else:
assert hasattr(filename_or_file, 'read'), 'expected file or str, got %s' % filename_or_file
self.file = filename_or_file
self.own_file = False
def writekvs(self, kvs):
key2str = {}
for (key, val) in sorted(kvs.items()):
if isinstance(val, float):
valstr = '%-8.3g' % (val,)
else:
valstr = str(val)
key2str[self._truncate(key)] = self._truncate(valstr)
if len(key2str) == 0:
print('WARNING: tried to write empty key-value dict')
return
else:
keywidth = max(map(len, key2str.keys()))
valwidth = max(map(len, key2str.values()))
dashes = '-' * (keywidth + valwidth + 7)
lines = [dashes]
for (key, val) in sorted(key2str.items()):
lines.append('| %s%s | %s%s |' % (key, ' ' * (keywidth - len(key)), val, ' ' * (valwidth - len(val))))
lines.append(dashes)
self.file.write('\n'.join(lines) + '\n')
self.file.flush()
def _truncate(self, s):
return s[:20] + '...' if len(s) > 23 else s
def writeseq(self, seq):
seq = list(seq)
for (i, elem) in enumerate(seq):
self.file.write(elem)
if i < len(seq) - 1:
self.file.write(' ')
self.file.write('\n')
self.file.flush()
def close(self):
if self.own_file:
self.file.close()
class JSONOutputFormat(KVWriter):
def __init__(self, filename):
self.file = open(filename, 'wt')
def writekvs(self, kvs):
for (k, v) in sorted(kvs.items()):
if hasattr(v, 'dtype'):
v = v.tolist()
kvs[k] = float(v)
self.file.write(json.dumps(kvs) + '\n')
self.file.flush()
def close(self):
self.file.close()
class WandBOutputFormat(KVWriter):
def __init__(self, filename):
group = None
if filename.endswith('trial'):
group = filename[:-6]
wandb.init(project='web_drrn', name=filename, group=group)
def writekvs(self, kvs):
wandb.log(kvs)
def close(self):
pass
class CSVOutputFormat(KVWriter):
def __init__(self, filename):
self.file = open(filename, 'w+t')
self.keys = []
self.sep = ','
def writekvs(self, kvs):
extra_keys = kvs.keys() - self.keys
if extra_keys:
self.keys.extend(extra_keys)
self.file.seek(0)
lines = self.file.readlines()
self.file.seek(0)
for (i, k) in enumerate(self.keys):
if i > 0:
self.file.write(',')
self.file.write(k)
self.file.write('\n')
for line in lines[1:]:
self.file.write(line[:-1])
self.file.write(self.sep * len(extra_keys))
self.file.write('\n')
for (i, k) in enumerate(self.keys):
if i > 0:
self.file.write(',')
v = kvs.get(k)
if v is not None:
self.file.write(str(v))
self.file.write('\n')
self.file.flush()
def close(self):
self.file.close()
class TensorBoardOutputFormat(KVWriter):
def __init__(self, dir):
os.makedirs(dir, exist_ok=True)
self.dir = dir
self.step = 1
prefix = 'events'
path = osp.join(osp.abspath(dir), prefix)
import tensorflow as tf
from tensorflow.python import pywrap_tensorflow
from tensorflow.core.util import event_pb2
from tensorflow.python.util import compat
self.tf = tf
self.event_pb2 = event_pb2
self.pywrap_tensorflow = pywrap_tensorflow
self.writer = pywrap_tensorflow.EventsWriter(compat.as_bytes(path))
def writekvs(self, kvs):
def summary_val(k, v):
kwargs = {'tag': k, 'simple_value': float(v)}
return self.tf.Summary.Value(**kwargs)
summary = self.tf.Summary(value=[summary_val(k, v) for (k, v) in kvs.items()])
event = self.event_pb2.Event(wall_time=time.time(), summary=summary)
event.step = self.step
self.writer.WriteEvent(event)
self.writer.Flush()
self.step += 1
def close(self):
if self.writer:
self.writer.Close()
self.writer = None
def make_output_format(format, ev_dir, log_suffix='', args=None):
os.makedirs(ev_dir, exist_ok=True)
if format == 'stdout':
return HumanOutputFormat(sys.stdout)
elif format == 'log':
return HumanOutputFormat(osp.join(ev_dir, 'log%s.txt' % log_suffix))
elif format == 'json':
return JSONOutputFormat(osp.join(ev_dir, 'progress%s.json' % log_suffix))
elif format == 'csv':
return CSVOutputFormat(osp.join(ev_dir, 'progress%s.csv' % log_suffix))
elif format == 'tensorboard':
return TensorBoardOutputFormat(osp.join(ev_dir, 'tb%s' % log_suffix))
elif format == 'wandb':
return WandBOutputFormat(ev_dir)
else:
raise ValueError('Unknown format specified: %s' % (format,))
def logkv(key, val):
Logger.CURRENT.logkv(key, val)
def logkv_mean(key, val):
Logger.CURRENT.logkv_mean(key, val)
def logkvs(d):
for (k, v) in d.items():
logkv(k, v)
def dumpkvs():
Logger.CURRENT.dumpkvs()
def getkvs():
return Logger.CURRENT.name2val
def log(*args, level=INFO):
Logger.CURRENT.log(*args, level=level)
def debug(*args):
log(*args, level=DEBUG)
def info(*args):
log(*args, level=INFO)
def warn(*args):
log(*args, level=WARN)
def error(*args):
log(*args, level=ERROR)
def set_level(level):
Logger.CURRENT.set_level(level)
def get_dir():
return Logger.CURRENT.get_dir()
record_tabular = logkv
dump_tabular = dumpkvs
class ProfileKV:
def __init__(self, n):
self.n = 'wait_' + n
def __enter__(self):
self.t1 = time.time()
def __exit__(self, type, value, traceback):
Logger.CURRENT.name2val[self.n] += time.time() - self.t1
def profile(n):
def decorator_with_name(func):
def func_wrapper(*args, **kwargs):
with ProfileKV(n):
return func(*args, **kwargs)
return func_wrapper
return decorator_with_name
class Logger(object):
DEFAULT = None
CURRENT = None
def __init__(self, dir, output_formats):
self.name2val = defaultdict(float)
self.name2cnt = defaultdict(int)
self.level = INFO
self.dir = dir
self.output_formats = output_formats
def logkv(self, key, val):
self.name2val[key] = val
def logkv_mean(self, key, val):
if val is None:
self.name2val[key] = None
return
(oldval, cnt) = (self.name2val[key], self.name2cnt[key])
self.name2val[key] = oldval * cnt / (cnt + 1) + val / (cnt + 1)
self.name2cnt[key] = cnt + 1
def dumpkvs(self):
if self.level == DISABLED:
return
for fmt in self.output_formats:
if isinstance(fmt, KVWriter):
fmt.writekvs(self.name2val)
self.name2val.clear()
self.name2cnt.clear()
def log(self, *args, level=INFO):
if self.level <= level:
self._do_log(args)
def set_level(self, level):
self.level = level
def get_dir(self):
return self.dir
def close(self):
for fmt in self.output_formats:
fmt.close()
def _do_log(self, args):
for fmt in self.output_formats:
if isinstance(fmt, SeqWriter):
fmt.writeseq(map(str, args))
def configure(dir=None, format_strs=None):
if dir is None:
dir = os.getenv('OPENAI_LOGDIR')
if dir is None:
dir = osp.join(tempfile.gettempdir(), datetime.datetime.now().strftime('openai-%Y-%m-%d-%H-%M-%S-%f'))
assert isinstance(dir, str)
os.makedirs(dir, exist_ok=True)
log_suffix = ''
rank = 0
for varname in ['PMI_RANK', 'OMPI_COMM_WORLD_RANK']:
if varname in os.environ:
rank = int(os.environ[varname])
if rank > 0:
log_suffix = '-rank%03i' % rank
if format_strs is None:
if rank == 0:
format_strs = os.getenv('OPENAI_LOG_FORMAT', 'stdout,log,csv').split(',')
else:
format_strs = os.getenv('OPENAI_LOG_FORMAT_MPI', 'log').split(',')
format_strs = filter(None, format_strs)
output_formats = [make_output_format(f, dir, log_suffix) for f in format_strs]
Logger.CURRENT = Logger(dir=dir, output_formats=output_formats)
log('Logging to %s' % dir)
def _configure_default_logger():
format_strs = None
if 'OPENAI_LOG_FORMAT' not in os.environ:
format_strs = ['stdout']
configure(format_strs=format_strs)
Logger.DEFAULT = Logger.CURRENT
def reset():
if Logger.CURRENT is not Logger.DEFAULT:
Logger.CURRENT.close()
Logger.CURRENT = Logger.DEFAULT
log('Reset logger')
class scoped_configure(object):
def __init__(self, dir=None, format_strs=None):
self.dir = dir
self.format_strs = format_strs
self.prevlogger = None
def __enter__(self):
self.prevlogger = Logger.CURRENT
configure(dir=self.dir, format_strs=self.format_strs)
def __exit__(self, *args):
Logger.CURRENT.close()
Logger.CURRENT = self.prevlogger
def _demo():
info('hi')
debug("shouldn't appear")
set_level(DEBUG)
debug('should appear')
dir = '/tmp/testlogging'
if os.path.exists(dir):
shutil.rmtree(dir)
configure(dir=dir)
logkv('a', 3)
logkv('b', 2.5)
dumpkvs()
logkv('b', -2.5)
logkv('a', 5.5)
dumpkvs()
info('^^^ should see a = 5.5')
logkv_mean('b', -22.5)
logkv_mean('b', -44.4)
logkv('a', 5.5)
dumpkvs()
info('^^^ should see b = 33.3')
logkv('b', -2.5)
dumpkvs()
logkv('a', 'longasslongasslongasslongasslongasslongassvalue')
dumpkvs()
def read_json(fname):
import pandas
ds = []
with open(fname, 'rt') as fh:
for line in fh:
ds.append(json.loads(line))
return pandas.DataFrame(ds)
def read_csv(fname):
import pandas
return pandas.read_csv(fname, index_col=None, comment='#')
def read_tb(path):
import pandas
import numpy as np
from glob import glob
from collections import defaultdict
import tensorflow as tf
if osp.isdir(path):
fnames = glob(osp.join(path, 'events.*'))
elif osp.basename(path).startswith('events.'):
fnames = [path]
else:
raise NotImplementedError('Expected tensorboard file or directory containing them. Got %s' % path)
tag2pairs = defaultdict(list)
maxstep = 0
for fname in fnames:
for summary in tf.train.summary_iterator(fname):
if summary.step > 0:
for v in summary.summary.value:
pair = (summary.step, v.simple_value)
tag2pairs[v.tag].append(pair)
maxstep = max(summary.step, maxstep)
data = np.empty((maxstep, len(tag2pairs)))
data[:] = np.nan
tags = sorted(tag2pairs.keys())
for (colidx, tag) in enumerate(tags):
pairs = tag2pairs[tag]
for (step, value) in pairs:
data[step - 1, colidx] = value
return pandas.DataFrame(data, columns=tags)
if __name__ == '__main__':
_demo()
# File: WebShop-master/baseline_models/models/bert.py
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import BertModel, BertConfig, PretrainedConfig, PreTrainedModel
from transformers.modeling_outputs import SequenceClassifierOutput
from .modules import EncoderRNN, BiAttention, get_aggregated
class BertConfigForWebshop(PretrainedConfig):
model_type = 'bert'
def __init__(self, pretrained_bert=True, image=False, **kwargs):
self.pretrained_bert = pretrained_bert
self.image = image
super().__init__(**kwargs)
class BertModelForWebshop(PreTrainedModel):
config_class = BertConfigForWebshop
def __init__(self, config):
super().__init__(config)
bert_config = BertConfig.from_pretrained('bert-base-uncased')
if config.pretrained_bert:
self.bert = BertModel.from_pretrained('bert-base-uncased')
else:
self.bert = BertModel(config)
self.bert.resize_token_embeddings(30526)
self.attn = BiAttention(768, 0.0)
self.linear_1 = nn.Linear(768 * 4, 768)
self.relu = nn.ReLU()
self.linear_2 = nn.Linear(768, 1)
if config.image:
self.image_linear = nn.Linear(512, 768)
else:
self.image_linear = None
self.linear_3 = nn.Sequential(nn.Linear(768, 128), nn.LeakyReLU(), nn.Linear(128, 1))
def forward(self, state_input_ids, state_attention_mask, action_input_ids, action_attention_mask, sizes, images=None, labels=None):
sizes = sizes.tolist()
state_rep = self.bert(state_input_ids, attention_mask=state_attention_mask)[0]
if images is not None and self.image_linear is not None:
images = self.image_linear(images)
state_rep = torch.cat([images.unsqueeze(1), state_rep], dim=1)
state_attention_mask = torch.cat([state_attention_mask[:, :1], state_attention_mask], dim=1)
action_rep = self.bert(action_input_ids, attention_mask=action_attention_mask)[0]
state_rep = torch.cat([state_rep[i:i + 1].repeat(j, 1, 1) for (i, j) in enumerate(sizes)], dim=0)
state_attention_mask = torch.cat([state_attention_mask[i:i + 1].repeat(j, 1) for (i, j) in enumerate(sizes)], dim=0)
act_lens = action_attention_mask.sum(1).tolist()
state_action_rep = self.attn(action_rep, state_rep, state_attention_mask)
state_action_rep = self.relu(self.linear_1(state_action_rep))
act_values = get_aggregated(state_action_rep, act_lens, 'mean')
act_values = self.linear_2(act_values).squeeze(1)
logits = [F.log_softmax(_, dim=0) for _ in act_values.split(sizes)]
loss = None
if labels is not None:
loss = -sum([logit[label] for (logit, label) in zip(logits, labels)]) / len(logits)
return SequenceClassifierOutput(loss=loss, logits=logits)
def rl_forward(self, state_batch, act_batch, value=False, q=False, act=False):
act_values = []
act_sizes = []
values = []
for (state, valid_acts) in zip(state_batch, act_batch):
with torch.set_grad_enabled(not act):
state_ids = torch.tensor([state.obs]).cuda()
state_mask = (state_ids > 0).int()
act_lens = [len(_) for _ in valid_acts]
act_ids = [torch.tensor(_) for _ in valid_acts]
act_ids = nn.utils.rnn.pad_sequence(act_ids, batch_first=True).cuda()
act_mask = (act_ids > 0).int()
act_size = torch.tensor([len(valid_acts)]).cuda()
if self.image_linear is not None:
images = [state.image_feat]
images = [torch.zeros(512) if _ is None else _ for _ in images]
images = torch.stack(images).cuda()
else:
images = None
logits = self.forward(state_ids, state_mask, act_ids, act_mask, act_size, images=images).logits[0]
act_values.append(logits)
act_sizes.append(len(valid_acts))
if value:
v = self.bert(state_ids, state_mask)[0]
values.append(self.linear_3(v[0][0]))
act_values = torch.cat(act_values, dim=0)
act_values = torch.cat([F.log_softmax(_, dim=0) for _ in act_values.split(act_sizes)], dim=0)
if value:
values = torch.cat(values, dim=0)
return (act_values, act_sizes, values)
else:
return (act_values, act_sizes)
# File: WebShop-master/baseline_models/models/modules.py
import itertools
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn.utils import rnn
def duplicate(output, mask, lens, act_sizes):
output = torch.cat([output[i:i + 1].repeat(j, 1, 1) for (i, j) in enumerate(act_sizes)], dim=0)
mask = torch.cat([mask[i:i + 1].repeat(j, 1) for (i, j) in enumerate(act_sizes)], dim=0)
lens = list(itertools.chain.from_iterable([lens[i:i + 1] * j for (i, j) in enumerate(act_sizes)]))
return (output, mask, lens)
def get_aggregated(output, lens, method):
if method == 'mean':
return torch.stack([output[i, :j, :].mean(0) for (i, j) in enumerate(lens)], dim=0)
elif method == 'last':
return torch.stack([output[i, j - 1, :] for (i, j) in enumerate(lens)], dim=0)
elif method == 'first':
return output[:, 0, :]
class EncoderRNN(nn.Module):
def __init__(self, input_size, num_units, nlayers, concat, bidir, layernorm, return_last):
super().__init__()
self.layernorm = layernorm == 'layer'
if layernorm:
self.norm = nn.LayerNorm(input_size)
self.rnns = []
for i in range(nlayers):
if i == 0:
input_size_ = input_size
output_size_ = num_units
else:
input_size_ = num_units if not bidir else num_units * 2
output_size_ = num_units
self.rnns.append(nn.GRU(input_size_, output_size_, 1, bidirectional=bidir, batch_first=True))
self.rnns = nn.ModuleList(self.rnns)
self.init_hidden = nn.ParameterList([nn.Parameter(torch.zeros(size=(2 if bidir else 1, 1, num_units)), requires_grad=True) for _ in range(nlayers)])
self.concat = concat
self.nlayers = nlayers
self.return_last = return_last
self.reset_parameters()
def reset_parameters(self):
with torch.no_grad():
for rnn_layer in self.rnns:
for (name, p) in rnn_layer.named_parameters():
if 'weight_ih' in name:
torch.nn.init.xavier_uniform_(p.data)
elif 'weight_hh' in name:
torch.nn.init.orthogonal_(p.data)
elif 'bias' in name:
p.data.fill_(0.0)
else:
p.data.normal_(std=0.1)
def get_init(self, bsz, i):
return self.init_hidden[i].expand(-1, bsz, -1).contiguous()
def forward(self, inputs, input_lengths=None):
(bsz, slen) = (inputs.size(0), inputs.size(1))
if self.layernorm:
inputs = self.norm(inputs)
output = inputs
outputs = []
lens = 0
if input_lengths is not None:
lens = input_lengths
for i in range(self.nlayers):
hidden = self.get_init(bsz, i)
if input_lengths is not None:
output = rnn.pack_padded_sequence(output, lens, batch_first=True, enforce_sorted=False)
(output, hidden) = self.rnns[i](output, hidden)
if input_lengths is not None:
(output, _) = rnn.pad_packed_sequence(output, batch_first=True)
if output.size(1) < slen:
padding = torch.zeros(size=(1, 1, 1), dtype=output.type(), device=output.device())
output = torch.cat([output, padding.expand(output.size(0), slen - output.size(1), output.size(2))], dim=1)
if self.return_last:
outputs.append(hidden.permute(1, 0, 2).contiguous().view(bsz, -1))
else:
outputs.append(output)
if self.concat:
return torch.cat(outputs, dim=2)
return outputs[-1]
class BiAttention(nn.Module):
def __init__(self, input_size, dropout):
super().__init__()
self.dropout = nn.Dropout(dropout)
self.input_linear = nn.Linear(input_size, 1, bias=False)
self.memory_linear = nn.Linear(input_size, 1, bias=False)
self.dot_scale = nn.Parameter(torch.zeros(size=(input_size,)).uniform_(1.0 / input_size ** 0.5), requires_grad=True)
self.init_parameters()
def init_parameters(self):
return
def forward(self, context, memory, mask):
(bsz, input_len) = (context.size(0), context.size(1))
memory_len = memory.size(1)
context = self.dropout(context)
memory = self.dropout(memory)
input_dot = self.input_linear(context)
memory_dot = self.memory_linear(memory).view(bsz, 1, memory_len)
cross_dot = torch.bmm(context * self.dot_scale, memory.permute(0, 2, 1).contiguous())
att = input_dot + memory_dot + cross_dot
att = att - 1e+30 * (1 - mask[:, None])
weight_one = F.softmax(att, dim=-1)
output_one = torch.bmm(weight_one, memory)
weight_two = F.softmax(att.max(dim=-1)[0], dim=-1).view(bsz, 1, input_len)
output_two = torch.bmm(weight_two, context)
return torch.cat([context, output_one, context * output_one, output_two * output_one], dim=-1)
# File: WebShop-master/baseline_models/models/rnn.py
import torch
import torch.nn as nn
import torch.nn.functional as F
from .modules import EncoderRNN, BiAttention, get_aggregated, duplicate
class RCDQN(nn.Module):
def __init__(self, vocab_size, embedding_dim, hidden_dim, arch, grad, embs=None, gru_embed='embedding', get_image=0, bert_path=''):
super().__init__()
self.word_dim = embedding_dim
self.word_emb = nn.Embedding(vocab_size, embedding_dim)
if embs is not None:
print('Loading embeddings of shape {}'.format(embs.shape))
self.word_emb.weight.data.copy_(torch.from_numpy(embs))
self.hidden_dim = hidden_dim
self.keep_prob = 1.0
self.rnn = EncoderRNN(self.word_dim, self.hidden_dim, 1, concat=True, bidir=True, layernorm='None', return_last=False)
self.att_1 = BiAttention(self.hidden_dim * 2, 1 - self.keep_prob)
self.att_2 = BiAttention(self.hidden_dim * 2, 1 - self.keep_prob)
self.att_3 = BiAttention(embedding_dim, 1 - self.keep_prob)
self.linear_1 = nn.Sequential(nn.Linear(self.hidden_dim * 8, self.hidden_dim), nn.LeakyReLU())
self.rnn_2 = EncoderRNN(self.hidden_dim, self.hidden_dim, 1, concat=True, bidir=True, layernorm='layer', return_last=False)
self.linear_2 = nn.Sequential(nn.Linear(self.hidden_dim * 12, self.hidden_dim * 2), nn.LeakyReLU())
self.linear_3 = nn.Sequential(nn.Linear(self.hidden_dim * 2, self.hidden_dim), nn.LeakyReLU(), nn.Linear(self.hidden_dim, 1))
self.get_image = get_image
if self.get_image:
self.linear_image = nn.Linear(512, self.hidden_dim)
def prepare(self, ids):
lens = [len(_) for _ in ids]
ids = [torch.tensor(_) for _ in ids]
ids = nn.utils.rnn.pad_sequence(ids, batch_first=True).cuda()
mask = (ids > 0).float()
embed = self.word_emb(ids)
output = self.rnn(embed, lens)
return (ids, lens, mask, embed, output)
def forward(self, state_batch, act_batch, value=False, q=False, act=False):
if self.arch == 'bert':
return self.bert_forward(state_batch, act_batch, value, q, act)
(obs_ids, obs_lens, obs_mask, obs_embed, obs_output) = self.prepare([state.obs for state in state_batch])
(goal_ids, goal_lens, goal_mask, goal_embed, goal_output) = self.prepare([state.goal for state in state_batch])
state_output = self.att_1(obs_output, goal_output, goal_mask)
state_output = self.linear_1(state_output)
if self.get_image:
images = [state.image_feat for state in state_batch]
images = [torch.zeros(512) if _ is None else _ for _ in images]
images = torch.stack([_ for _ in images]).cuda()
images = self.linear_image(images)
state_output = torch.cat([images.unsqueeze(1), state_output], dim=1)
obs_lens = [_ + 1 for _ in obs_lens]
obs_mask = torch.cat([obs_mask[:, :1], obs_mask], dim=1)
state_output = self.rnn_2(state_output, obs_lens)
if value:
values = get_aggregated(state_output, obs_lens, 'mean')
values = self.linear_3(values).squeeze(1)
act_sizes = [len(_) for _ in act_batch]
act_batch = list(itertools.chain.from_iterable(act_batch))
(act_ids, act_lens, act_mask, act_embed, act_output) = self.prepare(act_batch)
(state_output, state_mask, state_lens) = duplicate(state_output, obs_mask, obs_lens, act_sizes)
(goal_embed, goal_mask, goal_lens) = duplicate(goal_embed, goal_mask, goal_lens, act_sizes)
state_act_output = self.att_2(act_output, state_output, state_mask)
goal_act_output = self.att_3(act_embed, goal_embed, goal_mask)
output = torch.cat([state_act_output, goal_act_output], dim=-1)
output = get_aggregated(output, act_lens, 'mean')
output = self.linear_2(output)
act_values = self.linear_3(output).squeeze(1)
if not q:
act_values = torch.cat([F.log_softmax(_, dim=0) for _ in act_values.split(act_sizes)], dim=0)
if value:
return (act_values, act_sizes, values)
else:
return (act_values, act_sizes)
# File: WebShop-master/baseline_models/train_choice_il.py
""""""
import argparse
import json
import logging
import math
import os
import random
from pathlib import Path
import datasets
import torch
from datasets import load_dataset, load_metric
from torch.utils.data import DataLoader
from tqdm.auto import tqdm
import transformers
from accelerate import Accelerator
from accelerate.logging import get_logger
from accelerate.utils import set_seed
from huggingface_hub import Repository
from transformers import AdamW, AutoConfig, AutoModelForSequenceClassification, AutoTokenizer, BertModel, BertConfig, DataCollatorWithPadding, PretrainedConfig, PreTrainedModel, SchedulerType, default_data_collator, get_scheduler
from transformers.utils.versions import require_version
from datasets import Dataset
from transformers.modeling_outputs import SequenceClassifierOutput
import torch.nn as nn
import torch.nn.functional as F
import wandb
from models.bert import BertModelForWebshop, BertConfigForWebshop
logger = get_logger(__name__)
require_version('datasets>=1.8.0', 'To fix: pip install -r examples/pytorch/text-classification/requirements.txt')
task_to_keys = {'cola': ('sentence', None), 'mnli': ('premise', 'hypothesis'), 'mrpc': ('sentence1', 'sentence2'), 'qnli': ('question', 'sentence'), 'qqp': ('question1', 'question2'), 'rte': ('sentence1', 'sentence2'), 'sst2': ('sentence', None), 'stsb': ('sentence1', 'sentence2'), 'wnli': ('sentence1', 'sentence2')}
tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased', truncation_side='left')
print(len(tokenizer))
tokenizer.add_tokens(['[button]', '[button_]', '[clicked button]', '[clicked button_]'], special_tokens=True)
print(len(tokenizer))
PATH = './data/il_trajs_finalized_images.jsonl'
MEM_PATH = './data/il_trajs_mem_finalized_images.jsonl'
HUMAN_GOAL_PATH = './data/human_goals.json'
def process(s):
s = s.lower().replace('"', '').replace("'", '').strip()
s = s.replace('[sep]', '[SEP]')
return s
def process_goal(state):
state = state.lower().replace('"', '').replace("'", '')
state = state.replace('amazon shopping game\ninstruction:', '').replace('webshop\ninstruction:', '')
state = state.replace('\n[button] search [button_]', '').strip()
if ', and price lower than' in state:
state = state.split(', and price lower than')[0]
return state
def get_data(split, mem=False, filter_search=True):
path = MEM_PATH if mem else PATH
print('Loading data from {}'.format(path))
with open(path, 'r') as json_file:
json_list = list(json_file)
human_goals = json.load(open(HUMAN_GOAL_PATH, 'r'))
random.seed(233)
random.shuffle(json_list)
goal_range = range(len(human_goals))
if split == 'train':
goal_range = range(1500, len(human_goals))
elif split == 'eval':
goal_range = range(500, 1500)
elif split == 'test':
goal_range = range(0, 500)
bad = cnt = 0
(state_list, action_list, idx_list, size_list) = ([], [], [], [])
image_list = []
num_trajs = 0
for json_str in json_list:
result = json.loads(json_str)
s = process_goal(result['states'][0])
assert s in human_goals, s
goal_idx = human_goals.index(s)
if goal_idx not in goal_range:
continue
num_trajs += 1
if 'images' not in result:
result['images'] = [0] * len(result['states'])
for (state, valid_acts, idx, image) in zip(result['states'], result['available_actions'], result['action_idxs'], result['images']):
cnt += 1
if filter_search and idx == -1:
continue
state_list.append(state)
image_list.append([0.0] * 512 if image == 0 else image)
if len(valid_acts) > 20:
bad += 1
new_idxs = list(range(6)) + random.sample(range(6, len(valid_acts)), 10)
if idx not in new_idxs:
new_idxs += [idx]
new_idxs = sorted(new_idxs)
valid_acts = [valid_acts[i] for i in new_idxs]
idx = new_idxs.index(idx)
action_list.extend(valid_acts)
idx_list.append(idx)
size_list.append(len(valid_acts))
print('num of {} trajs: {}'.format(split, num_trajs))
print('total transitions and bad transitions: {} {}'.format(cnt, bad))
(state_list, action_list) = (list(map(process, state_list)), list(map(process, action_list)))
return (state_list, action_list, idx_list, size_list, image_list)
def get_dataset(split, mem=False):
(states, actions, idxs, sizes, images) = get_data(split, mem)
state_encodings = tokenizer(states, padding='max_length', max_length=512, truncation=True, return_tensors='pt')
action_encodings = tokenizer(actions, padding='max_length', max_length=128, truncation=True, return_tensors='pt')
dataset = {'state_input_ids': state_encodings['input_ids'], 'state_attention_mask': state_encodings['attention_mask'], 'action_input_ids': action_encodings['input_ids'].split(sizes), 'action_attention_mask': action_encodings['attention_mask'].split(sizes), 'sizes': sizes, 'images': torch.tensor(images), 'labels': idxs}
return Dataset.from_dict(dataset)
def data_collator(batch):
(state_input_ids, state_attention_mask, action_input_ids, action_attention_mask, sizes, labels, images) = ([], [], [], [], [], [], [])
for sample in batch:
state_input_ids.append(sample['state_input_ids'])
state_attention_mask.append(sample['state_attention_mask'])
action_input_ids.extend(sample['action_input_ids'])
action_attention_mask.extend(sample['action_attention_mask'])
sizes.append(sample['sizes'])
labels.append(sample['labels'])
images.append(sample['images'])
max_state_len = max((sum(x) for x in state_attention_mask))
max_action_len = max((sum(x) for x in action_attention_mask))
return {'state_input_ids': torch.tensor(state_input_ids)[:, :max_state_len], 'state_attention_mask': torch.tensor(state_attention_mask)[:, :max_state_len], 'action_input_ids': torch.tensor(action_input_ids)[:, :max_action_len], 'action_attention_mask': torch.tensor(action_attention_mask)[:, :max_action_len], 'sizes': torch.tensor(sizes), 'images': torch.tensor(images), 'labels': torch.tensor(labels)}
def parse_args():
parser = argparse.ArgumentParser(description='Finetune a transformers model on a text classification task')
parser.add_argument('--task_name', type=str, default='mprc', help='The name of the glue task to train on.', choices=list(task_to_keys.keys()))
parser.add_argument('--train_file', type=str, default=None, help='A csv or a json file containing the training data.')
parser.add_argument('--validation_file', type=str, default=None, help='A csv or a json file containing the validation data.')
parser.add_argument('--max_length', type=int, default=128, help='The maximum total input sequence length after tokenization. Sequences longer than this will be truncated, sequences shorter will be padded if `--pad_to_max_lengh` is passed.')
parser.add_argument('--pad_to_max_length', action='store_true', help='If passed, pad all samples to `max_length`. Otherwise, dynamic padding is used.')
parser.add_argument('--model_name_or_path', default='bert-base-uncased', type=str, help='Path to pretrained model or model identifier from huggingface.co/models.')
parser.add_argument('--use_slow_tokenizer', action='store_true', help='If passed, will use a slow tokenizer (not backed by the 🤗 Tokenizers library).')
parser.add_argument('--per_device_train_batch_size', type=int, default=1, help='Batch size (per device) for the training dataloader.')
parser.add_argument('--per_device_eval_batch_size', type=int, default=8, help='Batch size (per device) for the evaluation dataloader.')
parser.add_argument('--learning_rate', type=float, default=2e-05, help='Initial learning rate (after the potential warmup period) to use.')
parser.add_argument('--weight_decay', type=float, default=0.0, help='Weight decay to use.')
parser.add_argument('--num_train_epochs', type=int, default=10, help='Total number of training epochs to perform.')
parser.add_argument('--max_train_steps', type=int, default=None, help='Total number of training steps to perform. If provided, overrides num_train_epochs.')
parser.add_argument('--gradient_accumulation_steps', type=int, default=32, help='Number of updates steps to accumulate before performing a backward/update pass.')
parser.add_argument('--lr_scheduler_type', type=SchedulerType, default='linear', help='The scheduler type to use.', choices=['linear', 'cosine', 'cosine_with_restarts', 'polynomial', 'constant', 'constant_with_warmup'])
parser.add_argument('--num_warmup_steps', type=int, default=0, help='Number of steps for the warmup in the lr scheduler.')
parser.add_argument('--output_dir', type=str, default='./ckpts/web_click', help='Where to store the final model.')
parser.add_argument('--seed', type=int, default=None, help='A seed for reproducible training.')
parser.add_argument('--push_to_hub', action='store_true', help='Whether or not to push the model to the Hub.')
parser.add_argument('--hub_model_id', type=str, help='The name of the repository to keep in sync with the local `output_dir`.')
parser.add_argument('--hub_token', type=str, help='The token to use to push to the Model Hub.')
parser.add_argument('--checkpointing_steps', type=str, default='epoch', help="Whether the various states should be saved at the end of every n steps, or 'epoch' for each epoch.")
parser.add_argument('--resume_from_checkpoint', type=str, default=None, help='If the training should continue from a checkpoint folder.')
parser.add_argument('--with_tracking', type=int, default=1, help='Whether to load in all available experiment trackers from the environment and use them for logging.')
parser.add_argument('--mem', type=int, default=0, help='State with memory')
parser.add_argument('--image', type=int, default=1, help='State with image')
parser.add_argument('--pretrain', type=int, default=1, help='Pretrained BERT or not')
parser.add_argument('--logging_steps', type=int, default=10, help='Logging in training')
args = parser.parse_args()
if args.task_name is None and args.train_file is None and (args.validation_file is None):
raise ValueError('Need either a task name or a training/validation file.')
else:
if args.train_file is not None:
extension = args.train_file.split('.')[-1]
assert extension in ['csv', 'json'], '`train_file` should be a csv or a json file.'
if args.validation_file is not None:
extension = args.validation_file.split('.')[-1]
assert extension in ['csv', 'json'], '`validation_file` should be a csv or a json file.'
if args.push_to_hub:
assert args.output_dir is not None, 'Need an `output_dir` to create a repo when `--push_to_hub` is passed.'
return args
def main():
args = parse_args()
accelerator = Accelerator()
wandb.init(project='bert_il', config=args, name=args.output_dir)
logging.basicConfig(format='%(asctime)s - %(levelname)s - %(name)s - %(message)s', datefmt='%m/%d/%Y %H:%M:%S', level=logging.INFO)
logger.info(accelerator.state, main_process_only=False)
if accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
transformers.utils.logging.set_verbosity_info()
else:
datasets.utils.logging.set_verbosity_error()
transformers.utils.logging.set_verbosity_error()
if args.seed is not None:
set_seed(args.seed)
config = BertConfigForWebshop(image=args.image, pretrain_bert=args.pretrain)
model = BertModelForWebshop(config)
train_dataset = get_dataset('train', mem=args.mem)
eval_dataset = get_dataset('eval', mem=args.mem)
for index in random.sample(range(len(train_dataset)), 3):
logger.info(f'Sample {index} of the training set: {train_dataset[index]}.')
train_dataloader = DataLoader(train_dataset, shuffle=True, collate_fn=data_collator, batch_size=args.per_device_train_batch_size)
eval_dataloader = DataLoader(eval_dataset, collate_fn=data_collator, batch_size=args.per_device_eval_batch_size)
no_decay = ['bias', 'LayerNorm.weight']
optimizer_grouped_parameters = [{'params': [p for (n, p) in model.named_parameters() if not any((nd in n for nd in no_decay))], 'weight_decay': args.weight_decay}, {'params': [p for (n, p) in model.named_parameters() if any((nd in n for nd in no_decay))], 'weight_decay': 0.0}]
optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
if args.max_train_steps is None:
args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
else:
args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
lr_scheduler = get_scheduler(name=args.lr_scheduler_type, optimizer=optimizer, num_warmup_steps=args.num_warmup_steps, num_training_steps=args.max_train_steps)
(model, optimizer, train_dataloader, eval_dataloader, lr_scheduler) = accelerator.prepare(model, optimizer, train_dataloader, eval_dataloader, lr_scheduler)
num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
if hasattr(args.checkpointing_steps, 'isdigit'):
checkpointing_steps = args.checkpointing_steps
if args.checkpointing_steps.isdigit():
checkpointing_steps = int(args.checkpointing_steps)
else:
checkpointing_steps = None
if args.with_tracking:
experiment_config = vars(args)
experiment_config['lr_scheduler_type'] = experiment_config['lr_scheduler_type'].value
accelerator.init_trackers('glue_no_trainer', experiment_config)
metric = load_metric('accuracy')
total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
logger.info('***** Running training *****')
logger.info(f' Num examples = {len(train_dataset)}')
logger.info(f' Num Epochs = {args.num_train_epochs}')
logger.info(f' Instantaneous batch size per device = {args.per_device_train_batch_size}')
logger.info(f' Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}')
logger.info(f' Gradient Accumulation steps = {args.gradient_accumulation_steps}')
logger.info(f' Total optimization steps = {args.max_train_steps}')
progress_bar = tqdm(range(args.max_train_steps), disable=not accelerator.is_local_main_process)
completed_steps = 0
starting_epoch = 0
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != '':
accelerator.print(f'Resumed from checkpoint: {args.resume_from_checkpoint}')
accelerator.load_state(args.resume_from_checkpoint)
path = os.path.basename(args.resume_from_checkpoint)
else:
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1]
training_difference = os.path.splitext(path)[0]
if 'epoch' in training_difference:
starting_epoch = int(training_difference.replace('epoch_', '')) + 1
resume_step = None
else:
resume_step = int(training_difference.replace('step_', ''))
starting_epoch = resume_step // len(train_dataloader)
resume_step -= starting_epoch * len(train_dataloader)
for epoch in range(starting_epoch, args.num_train_epochs):
model.train()
if args.with_tracking:
total_loss = total_step = 0
for (step, batch) in enumerate(train_dataloader):
if args.resume_from_checkpoint and epoch == starting_epoch:
if resume_step is not None and step < resume_step:
completed_steps += 1
continue
outputs = model(**batch)
loss = outputs.loss
if args.with_tracking:
total_loss += loss.detach().float()
total_step += 1
loss = loss / args.gradient_accumulation_steps
accelerator.backward(loss)
metric.add_batch(predictions=torch.stack([logit.argmax(dim=0) for logit in outputs.logits]), references=batch['labels'])
if step % args.gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
optimizer.step()
lr_scheduler.step()
optimizer.zero_grad()
progress_bar.update(1)
completed_steps += 1
if args.with_tracking and args.logging_steps > 0 and (completed_steps % args.logging_steps == 0):
train_metric = metric.compute()
wandb.log({'train_accuracy': train_metric, 'train_loss': total_loss / total_step, 'train_step': completed_steps})
total_loss = total_step = 0
if isinstance(checkpointing_steps, int):
if completed_steps % checkpointing_steps == 0:
output_dir = f'step_{completed_steps}'
if args.output_dir is not None:
output_dir = os.path.join(args.output_dir, output_dir)
accelerator.save_state(output_dir)
if completed_steps >= args.max_train_steps:
break
model.eval()
samples_seen = 0
total_loss = total_step = 0
if len(metric) > 0:
metric.compute()
for (step, batch) in enumerate(eval_dataloader):
with torch.no_grad():
outputs = model(**batch)
predictions = torch.stack([logit.argmax(dim=0) for logit in outputs.logits])
(predictions, references) = accelerator.gather((predictions, batch['labels']))
if accelerator.num_processes > 1:
if step == len(eval_dataloader):
predictions = predictions[:len(eval_dataloader.dataset) - samples_seen]
references = references[:len(eval_dataloader.dataset) - samples_seen]
else:
samples_seen += references.shape[0]
metric.add_batch(predictions=predictions, references=references)
total_loss += outputs.loss.detach().float()
total_step += 1
eval_metric = metric.compute()
logger.info(f'epoch {epoch}: {eval_metric}')
if args.with_tracking:
wandb.log({'eval_accuracy': eval_metric, 'eval_loss': total_loss / total_step, 'epoch': epoch, 'epoch_step': completed_steps})
if args.checkpointing_steps == 'epoch':
output_dir = f'epoch_{epoch}'
if args.output_dir is not None:
output_dir = os.path.join(args.output_dir, output_dir)
os.makedirs(output_dir, exist_ok=True)
unwrapped_model = accelerator.unwrap_model(model)
torch.save(unwrapped_model.state_dict(), os.path.join(output_dir, 'model.pth'))
if args.output_dir is not None:
with open(os.path.join(args.output_dir, 'all_results.json'), 'w') as f:
json.dump({'eval_accuracy': eval_metric['accuracy']}, f)
if __name__ == '__main__':
main()
# File: WebShop-master/baseline_models/train_rl.py
import argparse
import logging
import time
import torch
from collections import defaultdict
import logger
from agent import Agent, TransitionPG
from env import WebEnv
logging.getLogger().setLevel(logging.CRITICAL)
def configure_logger(log_dir, wandb):
logger.configure(log_dir, format_strs=['log'])
global tb
type_strs = ['json', 'stdout']
if wandb:
type_strs += ['wandb']
tb = logger.Logger(log_dir, [logger.make_output_format(type_str, log_dir) for type_str in type_strs])
global log
log = logger.log
def evaluate(agent, env, split, nb_episodes=10):
with torch.no_grad():
total_score = 0
for method in ['greedy']:
for ep in range(nb_episodes):
log('Starting {} episode {}'.format(split, ep))
if split == 'eval':
score = evaluate_episode(agent, env, split, method)
elif split == 'test':
score = evaluate_episode(agent, env, split, method, idx=ep)
log('{} episode {} ended with score {}\n\n'.format(split, ep, score))
total_score += score
avg_score = total_score / nb_episodes
return avg_score
def evaluate_episode(agent, env, split, method='greedy', idx=None):
step = 0
done = False
(ob, info) = env.reset(idx)
state = agent.build_state(ob, info)
log('Obs{}: {}'.format(step, ob.encode('utf-8')))
while not done:
valid_acts = info['valid']
with torch.no_grad():
action_str = agent.act([state], [valid_acts], method=method)[0][0]
log('Action{}: {}'.format(step, action_str))
(ob, rew, done, info) = env.step(action_str)
log('Reward{}: {}, Score {}, Done {}'.format(step, rew, info['score'], done))
step += 1
log('Obs{}: {}'.format(step, ob.encode('utf-8')))
state = agent.build_state(ob, info)
tb.logkv_mean(f'{split}Score', info['score'])
if 'verbose' in info:
for (k, v) in info['verbose'].items():
if k.startswith('r'):
tb.logkv_mean(f'{split}_' + k, v)
return info['score']
def agg(envs, attr):
res = defaultdict(int)
for env in envs:
for (k, v) in getattr(env, attr).items():
res[k] += v
return res
def train(agent, eval_env, test_env, envs, args):
start = time.time()
(states, valids, transitions) = ([], [], [])
state0 = None
for env in envs:
(ob, info) = env.reset()
if state0 is None:
state0 = (ob, info)
states.append(agent.build_state(ob, info))
valids.append(info['valid'])
for step in range(1, args.max_steps + 1):
(action_strs, action_ids, values) = agent.act(states, valids, method=args.exploration_method)
with torch.no_grad():
(action_values, _) = agent.network.rl_forward(states[:1], agent.encode_valids(valids[:1]))
actions = sorted(zip(state0[1]['valid'], action_values.tolist()), key=lambda x: -x[1])
log('State {}: {}'.format(step, state0[0].lower().encode('utf-8')))
log('Goal {}: {}'.format(step, state0[1]['goal'].lower().encode('utf-8')))
log('Actions{}: {}'.format(step, actions))
log('>> Values{}: {}'.format(step, float(values[0])))
log('>> Action{}: {}'.format(step, action_strs[0]))
state0 = None
(next_states, next_valids, rewards, dones) = ([], [], [], [])
for (env, action_str, action_id, state) in zip(envs, action_strs, action_ids, states):
(ob, reward, done, info) = env.step(action_str)
if state0 is None:
state0 = (ob, info)
r_att = r_opt = 0
if 'verbose' in info:
r_att = info['verbose'].get('r_att', 0)
r_option = info['verbose'].get('r_option ', 0)
r_price = info['verbose'].get('r_price', 0)
r_type = info['verbose'].get('r_type', 0)
w_att = info['verbose'].get('w_att', 0)
w_option = info['verbose'].get('w_option', 0)
w_price = info['verbose'].get('w_price', 0)
reward_str = f'{reward / 10:.2f} = ({r_att:.2f} * {w_att:.2f} + {r_option:.2f} * {w_option:.2f} + {r_price:.2f} * {w_price:.2f}) * {r_type:.2f}'
else:
reward_str = str(reward)
log('Reward{}: {}, Done {}\n'.format(step, reward_str, done))
next_state = agent.build_state(ob, info)
next_valid = info['valid']
(next_states, next_valids, rewards, dones) = (next_states + [next_state], next_valids + [next_valid], rewards + [reward], dones + [done])
if done:
tb.logkv_mean('EpisodeScore', info['score'])
category = env.session['goal']['category']
tb.logkv_mean(f'EpisodeScore_{category}', info['score'])
if 'verbose' in info:
for (k, v) in info['verbose'].items():
if k.startswith('r'):
tb.logkv_mean(k, v)
transitions.append(TransitionPG(states, action_ids, rewards, values, agent.encode_valids(valids), dones))
if len(transitions) >= args.bptt:
(_, _, last_values) = agent.act(next_states, next_valids, method='softmax')
stats = agent.update(transitions, last_values, step=step)
for (k, v) in stats.items():
tb.logkv_mean(k, v)
del transitions[:]
torch.cuda.empty_cache()
for (i, env) in enumerate(envs):
if dones[i]:
(ob, info) = env.reset()
if i == 0:
state0 = (ob, info)
next_states[i] = agent.build_state(ob, info)
next_valids[i] = info['valid']
(states, valids) = (next_states, next_valids)
if step % args.eval_freq == 0:
evaluate(agent, eval_env, 'eval')
if step % args.test_freq == 0:
evaluate(agent, test_env, 'test', 500)
if step % args.log_freq == 0:
tb.logkv('Step', step)
tb.logkv('FPS', int(step * len(envs) / (time.time() - start)))
for (k, v) in agg(envs, 'stats').items():
tb.logkv(k, v)
items_clicked = agg(envs, 'items_clicked')
tb.logkv('ItemsClicked', len(items_clicked))
tb.dumpkvs()
if step % args.ckpt_freq == 0:
agent.save()
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--seed', default=0, type=int)
parser.add_argument('--output_dir', default='logs')
parser.add_argument('--ckpt_freq', default=10000, type=int)
parser.add_argument('--eval_freq', default=500, type=int)
parser.add_argument('--test_freq', default=5000, type=int)
parser.add_argument('--log_freq', default=100, type=int)
parser.add_argument('--wandb', default=1, type=int)
parser.add_argument('--num_envs', default=4, type=int)
parser.add_argument('--step_limit', default=100, type=int)
parser.add_argument('--max_steps', default=300000, type=int)
parser.add_argument('--learning_rate', default=1e-05, type=float)
parser.add_argument('--gamma', default=0.9, type=float)
parser.add_argument('--clip', default=10, type=float)
parser.add_argument('--bptt', default=8, type=int)
parser.add_argument('--exploration_method', default='softmax', type=str, choices=['eps', 'softmax'])
parser.add_argument('--w_pg', default=1, type=float)
parser.add_argument('--w_td', default=1, type=float)
parser.add_argument('--w_il', default=0, type=float)
parser.add_argument('--w_en', default=1, type=float)
parser.add_argument('--network', default='bert', type=str, choices=['bert', 'rnn'])
parser.add_argument('--bert_path', default='', type=str, help='which bert to load')
parser.add_argument('--embedding_dim', default=128, type=int)
parser.add_argument('--hidden_dim', default=128, type=int)
parser.add_argument('--grad_encoder', default=1, type=int)
parser.add_argument('--get_image', default=1, type=int, help='use image in models')
parser.add_argument('--num', default=None, type=int)
parser.add_argument('--click_item_name', default=1, type=int)
parser.add_argument('--state_format', default='text_rich', type=str)
parser.add_argument('--human_goals', default=1, type=int, help='use human goals')
parser.add_argument('--num_prev_obs', default=0, type=int, help='number of previous observations')
parser.add_argument('--num_prev_actions', default=0, type=int, help='number of previous actions')
parser.add_argument('--extra_search_path', default='./data/goal_query_predict.json', type=str, help='path for extra search queries')
parser.add_argument('--ban_buy', default=0, type=int, help='ban buy action before selecting options')
parser.add_argument('--score_handicap', default=0, type=int, help='provide score in state')
parser.add_argument('--go_to_item', default=0, type=int)
parser.add_argument('--go_to_search', default=0, type=int)
parser.add_argument('--harsh_reward', default=0, type=int)
parser.add_argument('--debug', default=0, type=int, help='debug mode')
parser.add_argument('--f', help='a dummy argument to fool ipython', default='1')
return parser.parse_known_args()
def main():
(args, unknown) = parse_args()
if args.debug:
args.num_envs = 2
args.wandb = 0
args.human_goals = 0
args.num = 100
print(unknown)
print(args)
configure_logger(args.output_dir, args.wandb)
agent = Agent(args)
train_env = WebEnv(args, split='train', id='train_')
server = train_env.env.server
eval_env = WebEnv(args, split='eval', id='eval_', server=server)
test_env = WebEnv(args, split='test', id='test_', server=server)
envs = [WebEnv(args, split='train', server=server, id=f'train{i}_') for i in range(args.num_envs)]
print('loaded')
train(agent, eval_env, test_env, envs, args)
if __name__ == '__main__':
main()
# File: WebShop-master/baseline_models/train_search_il.py
import json
import os
import random
from datasets import Dataset, DatasetDict, load_from_disk
from transformers import BartForConditionalGeneration, BartTokenizer, Trainer, TrainingArguments
from transformers.models.bart.modeling_bart import shift_tokens_right
tokenizer = BartTokenizer.from_pretrained('facebook/bart-large')
BOS_TOKEN_ID = 0
PAD_TOKEN_ID = 1
EOS_TOKEN_ID = 2
UNK_TOKEN_ID = 3
PATH = './data/goal_query_map.json'
HUMAN_GOAL_PATH = './data/human_goals.json'
GOAL_PATH = './data/items_human_ins.json'
def process_str(s):
s = s.lower().replace('"', '').replace("'", '').strip()
return s
def process_goal(state):
state = state.lower().replace('"', '').replace("'", '')
state = state.replace('amazon shopping game\ninstruction:', '').replace('webshop\ninstruction:', '')
state = state.replace('\n[button] search [button_]', '').strip()
if ', and price lower than' in state:
state = state.split(', and price lower than')[0]
return state
def get_data(split):
data = json.load(open(PATH))
(goals, searches) = ([], [])
for (goal, search_list) in data.items():
goal = process_goal(goal)
for search in search_list:
search = process_str(search)
goals.append(goal)
searches.append(search)
n = len(goals)
human_goals = json.load(open(HUMAN_GOAL_PATH, 'r'))
goal_range = range(len(human_goals))
if split == 'train':
goal_range = range(500, len(human_goals))
elif split == 'validation':
goal_range = range(500, 1500)
elif split == 'test':
goal_range = range(0, 500)
elif split == 'all':
all_data = json.load(open(GOAL_PATH))
all_goals = []
all_goals_processed = []
for ins_list in all_data.values():
for ins in ins_list:
ins = ins['instruction']
all_goals.append(ins)
all_goals_processed.append(process_str(ins))
return (all_goals_processed, all_goals)
(goals_, searches_) = ([], [])
for (goal, search) in zip(goals, searches):
if goal in human_goals and human_goals.index(goal) in goal_range:
goals_.append(goal)
searches_.append(search)
return (goals_, searches_)
def get_dataset(name, flip=False, variant=None, size=None):
fname = name + '-flip' if flip else name
fpath = os.path.join(os.path.dirname(__file__), fname)
d = {}
splits = ['train', 'validation', 'test']
if name == 'web_search':
splits = ['train', 'validation', 'test', 'all']
for split in splits:
(input, output) = get_data(split) if name != 'nl2bash' else get_data(split, variant=variant)
l = len(input) if size is None else int(len(input) * size)
print('{} size: {}'.format(split, l))
if flip:
(input, output) = (output, input)
(input, output) = (input[:l], output[:l])
d[split] = process_dataset(input, output)
d = DatasetDict(d)
return d
def process_dataset(input, output, max_len=256):
input_encodings = tokenizer(input, padding='max_length', max_length=max_len, truncation=True, return_tensors='pt')
output_encodings = tokenizer(output, padding='max_length', max_length=max_len, truncation=True, return_tensors='pt')
labels = output_encodings['input_ids']
decoder_input_ids = shift_tokens_right(labels, PAD_TOKEN_ID, EOS_TOKEN_ID)
labels[labels[:, :] == PAD_TOKEN_ID] = -100
dataset = Dataset.from_dict({'input_ids': input_encodings['input_ids'], 'attention_mask': input_encodings['attention_mask'], 'decoder_input_ids': decoder_input_ids, 'labels': labels})
dataset.set_format(type='torch', columns=['input_ids', 'labels', 'decoder_input_ids', 'attention_mask'])
return dataset
if __name__ == '__main__':
dataset = get_dataset('web_search', flip=False)
train_dataset = dataset['train']
print(train_dataset[0])
model = BartForConditionalGeneration.from_pretrained('facebook/bart-base')
model.resize_token_embeddings(len(tokenizer))
training_args = TrainingArguments(output_dir='./ckpts/web_search', num_train_epochs=10, per_device_train_batch_size=4, per_device_eval_batch_size=4, warmup_steps=50, weight_decay=0.01, evaluation_strategy='steps', logging_dir='./logs', logging_steps=50, eval_steps=20, save_steps=200)
trainer = Trainer(model=model, args=training_args, train_dataset=train_dataset, eval_dataset=dataset['validation'], compute_metrics=None)
trainer.train()
# File: WebShop-master/run_envs/run_web_agent_site_env.py
""""""
import gym
from rich import print
from rich.markup import escape
from web_agent_site.envs import WebAgentSiteEnv
from web_agent_site.models import HumanPolicy, RandomPolicy
from web_agent_site.utils import DEBUG_PROD_SIZE
if __name__ == '__main__':
env = WebAgentSiteEnv(observation_mode='text', render=False, num_products=DEBUG_PROD_SIZE)
global_step = 0
try:
policy = RandomPolicy()
observation = env.observation
while True:
print(observation)
available_actions = env.get_available_actions()
print('Available actions:', available_actions)
action = policy.forward(observation, available_actions)
(observation, reward, done, info) = env.step(action)
print(f'Taking action "{escape(action)}" -> Reward = {reward}')
if done:
break
global_step += 1
finally:
env.close()
# File: WebShop-master/run_envs/run_web_agent_text_env.py
""""""
import gym
from rich import print
from rich.markup import escape
from web_agent_site.envs import WebAgentTextEnv
from web_agent_site.models import RandomPolicy
from web_agent_site.utils import DEBUG_PROD_SIZE
if __name__ == '__main__':
env = gym.make('WebAgentTextEnv-v0', observation_mode='text', num_products=DEBUG_PROD_SIZE)
env.reset()
try:
policy = RandomPolicy()
observation = env.observation
while True:
print(observation)
available_actions = env.get_available_actions()
print('Available actions:', available_actions)
action = policy.forward(observation, available_actions)
(observation, reward, done, info) = env.step(action)
print(f'Taking action "{escape(action)}" -> Reward = {reward}')
if done:
break
finally:
env.close()
# File: WebShop-master/search_engine/convert_product_file_format.py
import sys
import json
from tqdm import tqdm
sys.path.insert(0, '../')
from web_agent_site.utils import DEFAULT_FILE_PATH
from web_agent_site.engine.engine import load_products
(all_products, *_) = load_products(filepath=DEFAULT_FILE_PATH)
docs = []
for p in tqdm(all_products, total=len(all_products)):
option_texts = []
options = p.get('options', {})
for (option_name, option_contents) in options.items():
option_contents_text = ', '.join(option_contents)
option_texts.append(f'{option_name}: {option_contents_text}')
option_text = ', and '.join(option_texts)
doc = dict()
doc['id'] = p['asin']
doc['contents'] = ' '.join([p['Title'], p['Description'], p['BulletPoints'][0], option_text]).lower()
doc['product'] = p
docs.append(doc)
with open('./resources_100/documents.jsonl', 'w+') as f:
for doc in docs[:100]:
f.write(json.dumps(doc) + '\n')
with open('./resources/documents.jsonl', 'w+') as f:
for doc in docs:
f.write(json.dumps(doc) + '\n')
with open('./resources_1k/documents.jsonl', 'w+') as f:
for doc in docs[:1000]:
f.write(json.dumps(doc) + '\n')
with open('./resources_100k/documents.jsonl', 'w+') as f:
for doc in docs[:100000]:
f.write(json.dumps(doc) + '\n')
# File: WebShop-master/search_engine/lucene_searcher.py
import json
from pyserini.search.lucene import LuceneSearcher
from rich import print
searcher = LuceneSearcher('indexes')
hits = searcher.search('rubber sole shoes', k=20)
for hit in hits:
doc = searcher.doc(hit.docid)
print(doc)
obj = json.loads(doc.raw())['product']['Title']
print(obj)
print(len(hits))
# File: WebShop-master/transfer/app.py
import gradio as gr
import json, time, torch
from transformers import BartTokenizer, BartForConditionalGeneration, AutoModel, AutoTokenizer
from webshop_lite import dict_to_fake_html
from predict_help import Page, convert_dict_to_actions, convert_html_to_text, parse_results_amz, parse_item_page_amz, parse_results_ws, parse_item_page_ws, parse_results_ebay, parse_item_page_ebay, WEBSHOP_URL, WEBSHOP_SESSION
ENVIRONMENTS = ['amazon', 'webshop', 'ebay']
BERT_MODEL_PATH = 'webshop/il-choice-bert-image_0'
bart_tokenizer = BartTokenizer.from_pretrained('facebook/bart-large')
bart_model = BartForConditionalGeneration.from_pretrained('webshop/il_search_bart')
bert_tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased', truncation_side='left')
bert_tokenizer.add_tokens(['[button]', '[button_]', '[clicked button]', '[clicked button_]'], special_tokens=True)
bert_model = AutoModel.from_pretrained(BERT_MODEL_PATH, trust_remote_code=True)
def process_str(s):
s = s.lower().replace('"', '').replace("'", '').strip()
s = s.replace('[sep]', '[SEP]')
return s
def process_goal(state):
state = state.lower().replace('"', '').replace("'", '')
state = state.replace('amazon shopping game\ninstruction:', '').replace('webshop\ninstruction:', '')
state = state.replace('\n[button] search [button_]', '').strip()
if ', and price lower than' in state:
state = state.split(', and price lower than')[0]
return state
def data_collator(batch):
(state_input_ids, state_attention_mask, action_input_ids, action_attention_mask, sizes, labels, images) = ([], [], [], [], [], [], [])
for sample in batch:
state_input_ids.append(sample['state_input_ids'])
state_attention_mask.append(sample['state_attention_mask'])
action_input_ids.extend(sample['action_input_ids'])
action_attention_mask.extend(sample['action_attention_mask'])
sizes.append(sample['sizes'])
labels.append(sample['labels'])
images.append(sample['images'])
max_state_len = max((sum(x) for x in state_attention_mask))
max_action_len = max((sum(x) for x in action_attention_mask))
return {'state_input_ids': torch.tensor(state_input_ids)[:, :max_state_len], 'state_attention_mask': torch.tensor(state_attention_mask)[:, :max_state_len], 'action_input_ids': torch.tensor(action_input_ids)[:, :max_action_len], 'action_attention_mask': torch.tensor(action_attention_mask)[:, :max_action_len], 'sizes': torch.tensor(sizes), 'images': torch.tensor(images), 'labels': torch.tensor(labels)}
def bart_predict(input):
input_ids = bart_tokenizer(input)['input_ids']
input_ids = torch.tensor(input_ids).unsqueeze(0)
output = bart_model.generate(input_ids, max_length=512, num_return_sequences=5, num_beams=5)
return bart_tokenizer.batch_decode(output.tolist(), skip_special_tokens=True)[0]
def bert_predict(obs, info, softmax=True):
valid_acts = info['valid']
assert valid_acts[0].startswith('click[')
state_encodings = bert_tokenizer(process_str(obs), max_length=512, truncation=True, padding='max_length')
action_encodings = bert_tokenizer(list(map(process_str, valid_acts)), max_length=512, truncation=True, padding='max_length')
batch = {'state_input_ids': state_encodings['input_ids'], 'state_attention_mask': state_encodings['attention_mask'], 'action_input_ids': action_encodings['input_ids'], 'action_attention_mask': action_encodings['attention_mask'], 'sizes': len(valid_acts), 'images': info['image_feat'].tolist(), 'labels': 0}
batch = data_collator([batch])
outputs = bert_model(**batch)
if softmax:
idx = torch.multinomial(torch.nn.functional.softmax(outputs.logits[0], dim=0), 1)[0].item()
else:
idx = outputs.logits[0].argmax(0).item()
return valid_acts[idx]
def get_return_value(env, asin, options, search_terms, page_num, product):
asin_url = None
if env == 'webshop':
query_str = '+'.join(search_terms.split())
options_str = json.dumps(options)
asin_url = f'{WEBSHOP_URL}/item_page/{WEBSHOP_SESSION}/{asin}/{query_str}/{page_num}/{options_str}'
else:
asin_url = f'https://www.ebay.com/itm/{asin}' if env == 'ebay' else f'https://www.amazon.com/dp/{asin}'
product_reduced = {k: v for (k, v) in product.items() if k in ['asin', 'Title', 'Description', 'BulletPoints']}
product_reduced['Description'] = product_reduced['Description'][:100] + '...'
product_reduced['Features'] = product_reduced.pop('BulletPoints')
product_reduced['Features'] = product_reduced['Features'][:100] + '...'
html = '<!DOCTYPE html><html><head><title>Chosen Product</title></head><body>'
html += f'''Product Image:<img src="{product['MainImage']}" height="50px" /><br>''' if len(product['MainImage']) > 0 else ''
html += f'Link to Product:\n <a href="{asin_url}" style="color:blue;text-decoration:underline;" target="_blank">{asin_url}</a>\n </body></html>'
return (product_reduced, options if len(options) > 0 else 'None Selected', html)
def predict(obs, info):
valid_acts = info['valid']
if valid_acts[0].startswith('click['):
return bert_predict(obs, info)
else:
return 'search[' + bart_predict(process_goal(obs)) + ']'
def run_episode(goal, env, verbose=True):
env = env.lower()
if env not in ENVIRONMENTS:
print(f'[ERROR] Environment {env} not recognized')
obs = 'Amazon Shopping Game\nInstruction:' + goal + '\n[button] search [button]'
info = {'valid': ['search[stuff]'], 'image_feat': torch.zeros(512)}
product_map = {}
title_to_asin_map = {}
search_results_cache = {}
(visited_asins, clicked_options) = (set(), set())
(sub_page_type, page_type, page_num) = (None, None, None)
(search_terms, prod_title, asin) = (None, None, None)
options = {}
for i in range(100):
action = predict(obs, info)
if verbose:
print('====')
print(action)
action_content = action[action.find('[') + 1:action.find(']')]
prev_page_type = page_type
if action.startswith('search['):
page_type = Page.RESULTS
search_terms = action_content
page_num = 1
elif action.startswith('click['):
if action.startswith('click[item -'):
prod_title = action_content[len('item -'):].strip()
found = False
for key in title_to_asin_map:
if prod_title == key:
asin = title_to_asin_map[key]
page_type = Page.ITEM_PAGE
visited_asins.add(asin)
found = True
break
if not found:
raise Exception('Product to click not found')
elif any((x.value in action for x in [Page.DESC, Page.FEATURES, Page.REVIEWS])):
page_type = Page.SUB_PAGE
sub_page_type = Page(action_content.lower())
elif action == 'click[< prev]':
if sub_page_type is not None:
(page_type, sub_page_type) = (Page.ITEM_PAGE, None)
elif prev_page_type == Page.ITEM_PAGE:
page_type = Page.RESULTS
(options, clicked_options) = ({}, set())
elif prev_page_type == Page.RESULTS and page_num > 1:
page_type = Page.RESULTS
page_num -= 1
elif action == 'click[next >]':
page_type = Page.RESULTS
page_num += 1
elif action.lower() == 'click[back to search]':
page_type = Page.SEARCH
elif action == 'click[buy now]':
return get_return_value(env, asin, options, search_terms, page_num, product_map[asin])
elif prev_page_type == Page.ITEM_PAGE:
found = False
for (opt_name, opt_values) in product_map[asin]['options'].items():
if action_content in opt_values:
options[opt_name] = action_content
page_type = Page.ITEM_PAGE
clicked_options.add(action_content)
found = True
break
if not found:
raise Exception('Unrecognized action: ' + action)
else:
raise Exception('Unrecognized action:' + action)
if verbose:
print(f'Parsing {page_type.value} page...')
if page_type == Page.RESULTS:
if search_terms in search_results_cache:
data = search_results_cache[search_terms]
if verbose:
print(f'Loading cached results page for "{search_terms}"')
else:
begin = time.time()
if env == 'amazon':
data = parse_results_amz(search_terms, page_num, verbose)
if env == 'webshop':
data = parse_results_ws(search_terms, page_num, verbose)
if env == 'ebay':
data = parse_results_ebay(search_terms, page_num, verbose)
end = time.time()
if verbose:
print(f'Parsing search results took {end - begin} seconds')
search_results_cache[search_terms] = data
for d in data:
title_to_asin_map[d['Title']] = d['asin']
elif page_type == Page.ITEM_PAGE or page_type == Page.SUB_PAGE:
if asin in product_map:
if verbose:
print('Loading cached item page for', asin)
data = product_map[asin]
else:
begin = time.time()
if env == 'amazon':
data = parse_item_page_amz(asin, verbose)
if env == 'webshop':
data = parse_item_page_ws(asin, search_terms, page_num, options, verbose)
if env == 'ebay':
data = parse_item_page_ebay(asin, verbose)
end = time.time()
if verbose:
print('Parsing item page took', end - begin, 'seconds')
product_map[asin] = data
elif page_type == Page.SEARCH:
if verbose:
print('Executing search')
obs = 'Amazon Shopping Game\nInstruction:' + goal + '\n[button] search [button]'
info = {'valid': ['search[stuff]'], 'image_feat': torch.zeros(512)}
continue
else:
raise Exception('Page of type `', page_type, '` not found')
begin = time.time()
html_str = dict_to_fake_html(data, page_type, asin, sub_page_type, options, product_map, goal)
obs = convert_html_to_text(html_str, simple=False, clicked_options=clicked_options, visited_asins=visited_asins)
end = time.time()
if verbose:
print('[Page Info -> WebShop HTML -> Observation] took', end - begin, 'seconds')
begin = time.time()
prod_arg = product_map if page_type == Page.ITEM_PAGE else data
info = convert_dict_to_actions(page_type, prod_arg, asin, page_num)
end = time.time()
if verbose:
print('Extracting available actions took', end - begin, 'seconds')
if i == 50:
return get_return_value(env, asin, options, search_terms, page_num, product_map[asin])
gr.Interface(fn=run_episode, inputs=[gr.inputs.Textbox(lines=7, label='Input Text'), gr.inputs.Radio(['Amazon', 'eBay'], type='value', default='Amazon', label='Environment')], outputs=[gr.outputs.JSON(label='Selected Product'), gr.outputs.JSON(label='Selected Options'), gr.outputs.HTML()], examples=[['I want to find a gold floor lamp with a glass shade and a nickel finish that i can use for my living room, and price lower than 270.00 dollars', 'Amazon'], ['I need some cute heart-shaped glittery cupcake picks as a gift to bring to a baby shower', 'Amazon'], ['I want to buy ballet shoes which have rubber sole in grey suede color and a size of 6', 'Amazon'], ['I would like a 7 piece king comforter set decorated with flowers and is machine washable', 'Amazon'], ["I'm trying to find white bluetooth speakers that are not only water resistant but also come with stereo sound", 'eBay'], ['find me the soy free 3.5 ounce 4-pack of dang thai rice chips, and make sure they are the aged cheddar flavor. i also need the ones in the resealable bags', 'eBay'], ['I am looking for a milk chocolate of 1 pound size in a single pack for valentine day', 'eBay'], ["I'm looking for a mini pc intel core desktop computer which supports with windows 11", 'eBay']], title='WebShop', article="<p style='padding-top:15px;text-align:center;'>To learn more about this project, check out the <a href='https://webshop-pnlp.github.io/' target='_blank'>project page</a>!</p>", description="<p style='text-align:center;'>Sim-to-real transfer of agent trained on WebShop to search a desired product on Amazon from any natural language query!</p>").launch(inline=False)
# File: WebShop-master/transfer/predict_help.py
from bs4 import BeautifulSoup
from bs4.element import Comment
from enum import Enum
import re, time
from urllib.parse import urlencode
import json, requests, torch
class Page(Enum):
DESC = 'description'
FEATURES = 'features'
ITEM_PAGE = 'item_page'
RESULTS = 'results'
REVIEWS = 'reviews'
SEARCH = 'search'
SUB_PAGE = 'item_sub_page'
HEADER_ = 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/101.0.4951.64 Safari/537.36'
DEBUG_HTML = 'temp.html'
NUM_PROD_LIMIT = 10
WEBSHOP_URL = 'http://3.83.245.205:3000'
WEBSHOP_SESSION = 'abc'
def parse_results_ebay(query, page_num=None, verbose=True):
query_string = '+'.join(query.split())
page_num = 1 if page_num is None else page_num
url = f'https://www.ebay.com/sch/i.html?_nkw={query_string}&_pgn={page_num}'
if verbose:
print(f'Search Results URL: {url}')
webpage = requests.get(url, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
soup = BeautifulSoup(webpage.text, 'html.parser')
products = soup.select('.s-item__wrapper.clearfix')
results = []
for item in products[:NUM_PROD_LIMIT]:
title = item.select_one('.s-item__title').text.strip()
if 'shop on ebay' in title.lower():
continue
link = item.select_one('.s-item__link')['href']
asin = link.split('?')[0][len('https://www.ebay.com/itm/'):]
try:
price = item.select_one('.s-item__price').text
if 'to' in price:
prices = price.split(' to ')
price = [p.strip('$') for p in prices]
except:
price = None
results.append({'asin': asin, 'Title': title, 'Price': price})
if verbose:
print(f'Scraped {len(results)} products')
return results
def parse_item_page_ebay(asin, verbose=True):
product_dict = {}
product_dict['asin'] = asin
url = f'https://www.ebay.com/itm/{asin}'
if verbose:
print(f'Item Page URL: {url}')
begin = time.time()
webpage = requests.get(url, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
end = time.time()
if verbose:
print(f'Item page scraping took {end - begin} seconds')
soup = BeautifulSoup(webpage.content, 'html.parser')
try:
product_dict['Title'] = soup.find('h1', {'class': 'x-item-title__mainTitle'}).text.strip()
except:
product_dict['Title'] = 'N/A'
try:
price_str = soup.find('div', {'class': 'mainPrice'}).text
prices = re.findall('\\d*\\.?\\d+', price_str)
product_dict['Price'] = prices[0]
except:
product_dict['Price'] = 'N/A'
try:
img_div = soup.find('div', {'id': 'mainImgHldr'})
img_link = img_div.find('img', {'id': 'icImg'})['src']
product_dict['MainImage'] = img_link
except:
product_dict['MainImage'] = ''
try:
rating = soup.find('span', {'class': 'reviews-star-rating'})['title'].split()[0]
except:
rating = None
product_dict['Rating'] = rating
(options, options_to_images) = ({}, {})
try:
option_blocks = soup.findAll('select', {'class': 'msku-sel'})
for block in option_blocks:
name = block['name'].strip().strip(':')
option_tags = block.findAll('option')
opt_list = []
for option_tag in option_tags:
if 'select' not in option_tag.text.lower():
opt_list.append(option_tag.text)
options[name] = opt_list
except:
options = {}
(product_dict['options'], product_dict['option_to_image']) = (options, options_to_images)
desc = None
try:
desc_link = soup.find('iframe', {'id': 'desc_ifr'})['src']
desc_webpage = requests.get(desc_link, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
desc_soup = BeautifulSoup(desc_webpage.content, 'html.parser')
desc = ' '.join(desc_soup.text.split())
except:
desc = 'N/A'
product_dict['Description'] = desc
features = None
try:
features = soup.find('div', {'class': 'x-about-this-item'}).text
except:
features = 'N/A'
product_dict['BulletPoints'] = features
return product_dict
def parse_results_ws(query, page_num=None, verbose=True):
query_string = '+'.join(query.split())
page_num = 1 if page_num is None else page_num
url = f'{WEBSHOP_URL}/search_results/{WEBSHOP_SESSION}/{query_string}/{page_num}'
if verbose:
print(f'Search Results URL: {url}')
webpage = requests.get(url, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
soup = BeautifulSoup(webpage.content, 'html.parser')
products = soup.findAll('div', {'class': 'list-group-item'})
results = []
for product in products:
asin = product.find('a', {'class': 'product-link'})
title = product.find('h4', {'class': 'product-title'})
price = product.find('h5', {'class': 'product-price'})
if '\n' in title:
title = title.text.split('\n')[0].strip()
else:
title = title.text.strip().strip('\n')
if 'to' in price.text:
prices = price.text.split(' to ')
price = [float(p.strip().strip('\n$')) for p in prices]
else:
price = float(price.text.strip().strip('\n$'))
results.append({'asin': asin.text, 'Title': title, 'Price': price})
if verbose:
print(f'Scraped {len(results)} products')
return results
def parse_item_page_ws(asin, query, page_num, options, verbose=True):
product_dict = {}
product_dict['asin'] = asin
query_string = '+'.join(query.split())
options_string = json.dumps(options)
url = f'{WEBSHOP_URL}/item_page/{WEBSHOP_SESSION}/{asin}/{query_string}/{page_num}/{options_string}'
if verbose:
print(f'Item Page URL: {url}')
webpage = requests.get(url, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
soup = BeautifulSoup(webpage.content, 'html.parser')
product_dict['Title'] = soup.find('h2').text
h4_headers = soup.findAll('h4')
for header in h4_headers:
text = header.text
if 'Price' in text:
product_dict['Price'] = text.split(':')[1].strip().strip('$')
elif 'Rating' in text:
product_dict['Rating'] = text.split(':')[1].strip()
product_dict['MainImage'] = soup.find('img')['src']
(options, options_to_image) = ({}, {})
option_blocks = soup.findAll('div', {'class': 'radio-toolbar'})
for block in option_blocks:
name = block.find('input')['name']
labels = block.findAll('label')
inputs = block.findAll('input')
opt_list = []
for (label, input) in zip(labels, inputs):
opt = label.text
opt_img_path = input['onclick'].split('href=')[1].strip("';")
opt_img_url = f'{WEBSHOP_URL}{opt_img_path}'
opt_list.append(opt)
options_to_image[opt] = opt_img_url
options[name] = opt_list
product_dict['options'] = options
product_dict['option_to_image'] = options_to_image
url = f'{WEBSHOP_URL}/item_sub_page/{WEBSHOP_SESSION}/{asin}/{query_string}/{page_num}/Description/{options_string}'
if verbose:
print(f'Item Description URL: {url}')
webpage = requests.get(url, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
soup = BeautifulSoup(webpage.content, 'html.parser')
product_dict['Description'] = soup.find(name='p', attrs={'class': 'product-info'}).text.strip()
url = f'{WEBSHOP_URL}/item_sub_page/{WEBSHOP_SESSION}/{asin}/{query_string}/{page_num}/Features/{options_string}'
if verbose:
print(f'Item Features URL: {url}')
webpage = requests.get(url, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
soup = BeautifulSoup(webpage.content, 'html.parser')
bullets = soup.find(name='ul').findAll(name='li')
product_dict['BulletPoints'] = '\n'.join([b.text.strip() for b in bullets])
return product_dict
def parse_results_amz(query, page_num=None, verbose=True):
url = 'https://www.amazon.com/s?k=' + query.replace(' ', '+')
if page_num is not None:
url += '&page=' + str(page_num)
if verbose:
print(f'Search Results URL: {url}')
webpage = requests.get(url, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
soup = BeautifulSoup(webpage.content, 'html.parser')
products = soup.findAll('div', {'data-component-type': 's-search-result'})
if products is None:
temp = open(DEBUG_HTML, 'w')
temp.write(str(soup))
temp.close()
raise Exception("Couldn't find search results page, outputted html for inspection")
results = []
for product in products[:NUM_PROD_LIMIT]:
asin = product['data-asin']
title = product.find('h2', {'class': 'a-size-mini'})
price_div = product.find('div', {'class': 's-price-instructions-style'})
price = price_div.find('span', {'class': 'a-offscreen'})
result = {'asin': asin, 'Title': title.text.strip(), 'Price': price.text.strip().strip('$')}
results.append(result)
if verbose:
print('Scraped', len(results), 'products')
return results
def parse_item_page_amz(asin, verbose=True):
product_dict = {}
product_dict['asin'] = asin
url = f'https://www.amazon.com/dp/{asin}'
if verbose:
print('Item Page URL:', url)
begin = time.time()
webpage = requests.get(url, headers={'User-Agent': HEADER_, 'Accept-Language': 'en-US, en;q=0.5'})
end = time.time()
if verbose:
print(f'Item page scraping took {end - begin} seconds')
soup = BeautifulSoup(webpage.content, 'html.parser')
try:
title = soup.find('span', attrs={'id': 'productTitle'})
title = title.string.strip().replace(',', '')
except AttributeError:
title = 'N/A'
product_dict['Title'] = title
try:
parent_price_span = soup.find(name='span', class_='apexPriceToPay')
price_span = parent_price_span.find(name='span', class_='a-offscreen')
price = float(price_span.getText().replace('$', ''))
except AttributeError:
price = 'N/A'
product_dict['Price'] = price
try:
rating = soup.find(name='span', attrs={'id': 'acrPopover'})
if rating is None:
rating = 'N/A'
else:
rating = rating.text
except AttributeError:
rating = 'N/A'
product_dict['Rating'] = rating.strip('\n').strip()
try:
features = soup.find(name='div', attrs={'id': 'feature-bullets'}).text
except AttributeError:
features = 'N/A'
product_dict['BulletPoints'] = features
try:
desc_body = soup.find(name='div', attrs={'id': 'productDescription_feature_div'})
desc_div = desc_body.find(name='div', attrs={'id': 'productDescription'})
desc_ps = desc_div.findAll(name='p')
desc = ' '.join([p.text for p in desc_ps])
except AttributeError:
desc = 'N/A'
product_dict['Description'] = desc.strip()
try:
imgtag = soup.find('img', {'id': 'landingImage'})
imageurl = dict(imgtag.attrs)['src']
except AttributeError:
imageurl = ''
product_dict['MainImage'] = imageurl
(options, options_to_image) = ({}, {})
try:
option_body = soup.find(name='div', attrs={'id': 'softlinesTwister_feature_div'})
if option_body is None:
option_body = soup.find(name='div', attrs={'id': 'twister_feature_div'})
option_blocks = option_body.findAll(name='ul')
for block in option_blocks:
name = json.loads(block['data-a-button-group'])['name']
opt_list = []
for li in block.findAll('li'):
img = li.find(name='img')
if img is not None:
opt = img['alt'].strip()
opt_img = img['src']
if len(opt) > 0:
options_to_image[opt] = opt_img
else:
opt = li.text.strip()
if len(opt) > 0:
opt_list.append(opt)
options[name.replace('_name', '').replace('twister_', '')] = opt_list
except AttributeError:
options = {}
(product_dict['options'], product_dict['option_to_image']) = (options, options_to_image)
return product_dict
def convert_html_to_text(html, simple=False, clicked_options=None, visited_asins=None):
def tag_visible(element):
ignore = {'style', 'script', 'head', 'title', 'meta', '[document]'}
return element.parent.name not in ignore and (not isinstance(element, Comment))
html_obj = BeautifulSoup(html, 'html.parser')
texts = html_obj.findAll(text=True)
visible_texts = filter(tag_visible, texts)
if simple:
return ' [SEP] '.join((t.strip() for t in visible_texts if t != '\n'))
else:
observation = ''
for t in visible_texts:
if t == '\n':
continue
if t.parent.name == 'button':
processed_t = f'[button] {t} [button]'
elif t.parent.name == 'label':
if f'{t}' in clicked_options:
processed_t = f' [clicked button] {t} [clicked button]'
observation = f'You have clicked {t}.\n' + observation
else:
processed_t = f' [button] {t} [button]'
elif t.parent.get('class') == ['product-link']:
if f'{t}' in visited_asins:
processed_t = f'\n[clicked button] {t} [clicked button]'
else:
processed_t = f'\n[button] {t} [button]'
else:
processed_t = str(t)
observation += processed_t + '\n'
return observation
def convert_dict_to_actions(page_type, products=None, asin=None, page_num=None) -> dict:
info = {'valid': []}
if page_type == Page.RESULTS:
info['valid'] = ['click[back to search]']
if products is None or page_num is None:
print(page_num)
print(products)
raise Exception('Provide `products`, `page_num` to get `results` valid actions')
if len(products) > 10:
info['valid'].append('click[next >]')
if page_num > 1:
info['valid'].append('click[< prev]')
for product in products:
info['valid'].append('click[item - ' + product['Title'] + ']')
if page_type == Page.ITEM_PAGE:
if products is None or asin is None:
raise Exception('Provide `products` and `asin` to get `item_page` valid actions')
info['valid'] = ['click[back to search]', 'click[< prev]', 'click[description]', 'click[features]', 'click[buy now]']
if 'options' in products[asin]:
for (key, values) in products[asin]['options'].items():
for value in values:
info['valid'].append('click[' + value + ']')
if page_type == Page.SUB_PAGE:
info['valid'] = ['click[back to search]', 'click[< prev]']
info['image_feat'] = torch.zeros(512)
return info
# File: WebShop-master/transfer/webshop_lite.py
import os
from flask import render_template_string, Flask
from predict_help import Page
app = Flask(__name__)
app.debug = True
SESSION_ID = 'ABC'
TEMPLATE_DIR = '../web_agent_site/templates/'
KEYWORDS = ['placeholder (not needed)']
QUERY = ''
product_map = {}
def read_html_template(path):
with open(path) as f:
template = f.read()
return template
@app.route('/', methods=['GET', 'POST'])
def index(session_id, **kwargs):
print('Hello world')
@app.route('/', methods=['GET', 'POST'])
def search_results(data):
path = os.path.join(TEMPLATE_DIR, 'results_page.html')
html = render_template_string(read_html_template(path=path), session_id=SESSION_ID, products=data, keywords=KEYWORDS, page=1, total=len(data), instruction_text=QUERY)
return html
@app.route('/', methods=['GET', 'POST'])
def item_page(session_id, asin, keywords, page, options):
path = os.path.join(TEMPLATE_DIR, 'item_page.html')
html = render_template_string(read_html_template(path=path), session_id=session_id, product_info=product_map[asin], keywords=keywords, page=page, asin=asin, options=options, instruction_text=QUERY)
return html
@app.route('/', methods=['GET', 'POST'])
def item_sub_page(session_id, asin, keywords, page, sub_page, options):
path = os.path.join(TEMPLATE_DIR, sub_page.value.lower() + '_page.html')
html = render_template_string(read_html_template(path), session_id=session_id, product_info=product_map[asin], keywords=keywords, page=page, asin=asin, options=options, instruction_text=QUERY)
return html
@app.route('/', methods=['GET', 'POST'])
def done(asin, options, session_id, **kwargs):
path = os.path.join(TEMPLATE_DIR, 'done_page.html')
html = render_template_string(read_html_template(path), session_id=session_id, reward=1, asin=asin, options=product_map[asin]['options'], reward_info=kwargs.get('reward_info'), goal_attrs=kwargs.get('goal_attrs'), purchased_attrs=kwargs.get('purchased_attrs'), goal=kwargs.get('goal'), mturk_code=kwargs.get('mturk_code'), query=kwargs.get('query'), category=kwargs.get('category'), product_category=kwargs.get('product_category'))
return html
def dict_to_fake_html(data, page_type, asin=None, sub_page_type=None, options=None, prod_map={}, query=''):
global QUERY, product_map
QUERY = query
product_map = prod_map
with app.app_context(), app.test_request_context():
if page_type == Page.RESULTS:
return search_results(data)
if page_type == Page.ITEM_PAGE:
return item_page(SESSION_ID, asin, KEYWORDS, 1, options)
if page_type == Page.SUB_PAGE:
if sub_page_type is not None:
return item_sub_page(SESSION_ID, asin, KEYWORDS, 1, sub_page_type, options)
else:
raise Exception('Sub page of type', sub_page_type, 'unrecognized')
# File: WebShop-master/web_agent_site/app.py
import argparse, json, logging, random
from pathlib import Path
from ast import literal_eval
from flask import Flask, request, redirect, url_for
from rich import print
from web_agent_site.engine.engine import load_products, init_search_engine, convert_web_app_string_to_var, get_top_n_product_from_keywords, get_product_per_page, map_action_to_html, END_BUTTON
from web_agent_site.engine.goal import get_reward, get_goals
from web_agent_site.utils import generate_mturk_code, setup_logger, DEFAULT_FILE_PATH, DEBUG_PROD_SIZE
app = Flask(__name__)
search_engine = None
all_products = None
product_item_dict = None
product_prices = None
attribute_to_asins = None
goals = None
weights = None
user_sessions = dict()
user_log_dir = None
SHOW_ATTRS_TAB = False
@app.route('/')
def home():
return redirect(url_for('index', session_id='abc'))
@app.route('/<session_id>', methods=['GET', 'POST'])
def index(session_id):
global user_log_dir
global all_products, product_item_dict, product_prices, attribute_to_asins, search_engine, goals, weights, user_sessions
if search_engine is None:
(all_products, product_item_dict, product_prices, attribute_to_asins) = load_products(filepath=DEFAULT_FILE_PATH, num_products=DEBUG_PROD_SIZE)
search_engine = init_search_engine(num_products=DEBUG_PROD_SIZE)
goals = get_goals(all_products, product_prices)
random.seed(233)
random.shuffle(goals)
weights = [goal['weight'] for goal in goals]
if session_id not in user_sessions and 'fixed' in session_id:
goal_dix = int(session_id.split('_')[-1])
goal = goals[goal_dix]
instruction_text = goal['instruction_text']
user_sessions[session_id] = {'goal': goal, 'done': False}
if user_log_dir is not None:
setup_logger(session_id, user_log_dir)
elif session_id not in user_sessions:
goal = random.choices(goals, weights)[0]
instruction_text = goal['instruction_text']
user_sessions[session_id] = {'goal': goal, 'done': False}
if user_log_dir is not None:
setup_logger(session_id, user_log_dir)
else:
instruction_text = user_sessions[session_id]['goal']['instruction_text']
if request.method == 'POST' and 'search_query' in request.form:
keywords = request.form['search_query'].lower().split(' ')
return redirect(url_for('search_results', session_id=session_id, keywords=keywords, page=1))
if user_log_dir is not None:
logger = logging.getLogger(session_id)
logger.info(json.dumps(dict(page='index', url=request.url, goal=user_sessions[session_id]['goal'])))
return map_action_to_html('start', session_id=session_id, instruction_text=instruction_text)
@app.route('/search_results/<session_id>/<keywords>/<page>', methods=['GET', 'POST'])
def search_results(session_id, keywords, page):
instruction_text = user_sessions[session_id]['goal']['instruction_text']
page = convert_web_app_string_to_var('page', page)
keywords = convert_web_app_string_to_var('keywords', keywords)
top_n_products = get_top_n_product_from_keywords(keywords, search_engine, all_products, product_item_dict, attribute_to_asins)
products = get_product_per_page(top_n_products, page)
html = map_action_to_html('search', session_id=session_id, products=products, keywords=keywords, page=page, total=len(top_n_products), instruction_text=instruction_text)
logger = logging.getLogger(session_id)
logger.info(json.dumps(dict(page='search_results', url=request.url, goal=user_sessions[session_id]['goal'], content=dict(keywords=keywords, search_result_asins=[p['asin'] for p in products], page=page))))
return html
@app.route('/item_page/<session_id>/<asin>/<keywords>/<page>/<options>', methods=['GET', 'POST'])
def item_page(session_id, asin, keywords, page, options):
options = literal_eval(options)
product_info = product_item_dict[asin]
goal_instruction = user_sessions[session_id]['goal']['instruction_text']
product_info['goal_instruction'] = goal_instruction
html = map_action_to_html('click', session_id=session_id, product_info=product_info, keywords=keywords, page=page, asin=asin, options=options, instruction_text=goal_instruction, show_attrs=SHOW_ATTRS_TAB)
logger = logging.getLogger(session_id)
logger.info(json.dumps(dict(page='item_page', url=request.url, goal=user_sessions[session_id]['goal'], content=dict(keywords=keywords, page=page, asin=asin, options=options))))
return html
@app.route('/item_sub_page/<session_id>/<asin>/<keywords>/<page>/<sub_page>/<options>', methods=['GET', 'POST'])
def item_sub_page(session_id, asin, keywords, page, sub_page, options):
options = literal_eval(options)
product_info = product_item_dict[asin]
goal_instruction = user_sessions[session_id]['goal']['instruction_text']
product_info['goal_instruction'] = goal_instruction
html = map_action_to_html(f'click[{sub_page}]', session_id=session_id, product_info=product_info, keywords=keywords, page=page, asin=asin, options=options, instruction_text=goal_instruction)
logger = logging.getLogger(session_id)
logger.info(json.dumps(dict(page='item_sub_page', url=request.url, goal=user_sessions[session_id]['goal'], content=dict(keywords=keywords, page=page, asin=asin, options=options))))
return html
@app.route('/done/<session_id>/<asin>/<options>', methods=['GET', 'POST'])
def done(session_id, asin, options):
options = literal_eval(options)
goal = user_sessions[session_id]['goal']
purchased_product = product_item_dict[asin]
price = product_prices[asin]
(reward, reward_info) = get_reward(purchased_product, goal, price=price, options=options, verbose=True)
user_sessions[session_id]['done'] = True
user_sessions[session_id]['reward'] = reward
print(user_sessions)
logger = logging.getLogger(session_id)
logger.info(json.dumps(dict(page='done', url=request.url, goal=goal, content=dict(asin=asin, options=options, price=price), reward=reward, reward_info=reward_info)))
del logging.root.manager.loggerDict[session_id]
return map_action_to_html(f'click[{END_BUTTON}]', session_id=session_id, reward=reward, asin=asin, options=options, reward_info=reward_info, query=purchased_product['query'], category=purchased_product['category'], product_category=purchased_product['product_category'], goal_attrs=user_sessions[session_id]['goal']['attributes'], purchased_attrs=purchased_product['Attributes'], goal=goal, mturk_code=generate_mturk_code(session_id))
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='WebShop flask app backend configuration')
parser.add_argument('--log', action='store_true', help='Log actions on WebShop in trajectory file')
parser.add_argument('--attrs', action='store_true', help='Show attributes tab in item page')
args = parser.parse_args()
if args.log:
user_log_dir = Path('user_session_logs/mturk')
user_log_dir.mkdir(parents=True, exist_ok=True)
SHOW_ATTRS_TAB = args.attrs
app.run(host='0.0.0.0', port=3000)
# File: WebShop-master/web_agent_site/attributes/annotate.py
import yaml
from pathlib import Path
from rich import print
ATTR_DIR = './data/attributes'
ATTR_PATHS = ['narrow_2-gram.yaml', 'narrow_1-gram.yaml', 'broad_2-gram.yaml', 'broad_1-gram.yaml']
ATTR_PATHS = [Path(ATTR_DIR) / af for af in ATTR_PATHS]
def annotate(attr_path):
with open(attr_path) as f:
attrs_by_cat = yaml.safe_load(f)
unique_attrs = set()
all_attrs = []
for (_, attrs) in attrs_by_cat.items():
attrs = [a.split('|')[0].strip() for a in attrs]
unique_attrs.update(attrs)
all_attrs += attrs
print(f'Total unique attributes: {len(unique_attrs)}')
total = len(all_attrs)
num_left = len(all_attrs)
annotated_attrs_by_cat = dict()
for (category, attrs) in attrs_by_cat.items():
print(f'Category: [ {category} ] | Number of attributes: {len(attrs)}\n')
annotated_attrs = []
for (i, attr) in enumerate(attrs):
(attr, score) = attr.split(' | ')
print(f"{'[' + str(i) + ']':<5} [bold green]{attr:<30}[/bold green] | [red]{category}[/red] | {score}")
tags = input('Annotate [1: ITEM, 2: PROP, 3: USE, ⎵: next example, q: next category] > ')
print('\n')
tags = tags.strip()
annotated_attrs.append(f'{attr} | {score} | {tags}')
if 'q' in tags:
break
num_left -= len(attrs)
print(f'{num_left} / {total} total attributes left.')
ans = input('Starting the next category... [y/n] > ')
if ans == 'n':
break
def main():
for attr_path in ATTR_PATHS:
annotate(attr_path)
if __name__ == '__main__':
''
main()
# File: WebShop-master/web_agent_site/attributes/generate_attrs.py
import json
import yaml
import random
from pathlib import Path
from collections import defaultdict
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.feature_extraction import text as sk_text
import pandas as pd
from tqdm import tqdm
from rich import print
ITEMS_PATH = './data/ITEMS_mar1.json'
REVIEWS_PATH = './data/reviews.json'
ATTR_DIR = './data/attributes'
random.seed(0)
def get_stop_words():
extra_stop_words = set([str(i) for i in range(1000)])
stop_words = sk_text.ENGLISH_STOP_WORDS.union(extra_stop_words)
return stop_words
def load_products(num=None):
with open(ITEMS_PATH) as f:
all_products = json.load(f)
if num is not None:
random.shuffle(all_products)
all_products = all_products[:num]
products = dict()
asins = set()
for p in all_products:
asin = p['asin']
if asin in asins:
continue
asins.add(asin)
products[asin] = p
with open(REVIEWS_PATH) as f:
reviews = json.load(f)
reviews = {r['asin']: r for r in reviews}
for (asin, p) in products.items():
if asin in reviews:
p['review'] = reviews[asin]
else:
p['review'] = None
return products
def get_top_attrs(attributes, k):
attr_to_asins = defaultdict(list)
for (asin, attr_scores) in attributes.items():
top_attr_scoress = attr_scores[:k]
for (attr, score) in top_attr_scoress:
attr_to_asins[attr].append(asin)
total = len([asin for (asin, _) in attributes.items()])
top_attrs = [(attr, len(asins) / total) for (attr, asins) in attr_to_asins.items()]
top_attrs = sorted(top_attrs, key=lambda x: -x[1])
top_attrs = [f'{attr} | {score:.4f}' for (attr, score) in top_attrs]
return top_attrs
def get_corpus(products, keys=('name', 'small_description'), category_type='category'):
all_products = list(products.values())
asins_by_cat = defaultdict(set)
corpus_by_cat = defaultdict(list)
for p in all_products:
category = p[category_type]
asin = p['asin']
if asin in asins_by_cat[category]:
continue
asins_by_cat[category].add(asin)
text = []
for key in keys:
if key == 'review':
rs = p['review']['reviews']
if r is not None:
text_ = ' '.join([r['review'].lower() for r in rs])
else:
text_ = ''
else:
text_ = p[key].lower()
text.append(text_)
text = ' '.join(text)
corpus_by_cat[category].append((asin, text))
return corpus_by_cat
def generate_ngram_attrs(corpus_by_cat, ngram_range, k, attrs):
vectorizer = TfidfVectorizer(stop_words=get_stop_words(), ngram_range=ngram_range, max_features=1000)
top_attrs_by_cat = dict()
for (category, corpus) in tqdm(corpus_by_cat.items(), total=len(corpus_by_cat)):
asins = [_[0] for _ in corpus]
texts = [_[1] for _ in corpus]
vec = vectorizer.fit_transform(texts).todense()
df = pd.DataFrame(vec, columns=vectorizer.get_feature_names_out())
attrs_by_cat = dict()
for (asin, (row_name, row)) in zip(asins, df.iterrows()):
attr_scores = sorted(list(zip(row.index, row)), key=lambda x: -x[1])
attrs_by_cat[asin] = attr_scores
attrs[asin] = attr_scores
top_attrs_by_cat[category.lower()] = get_top_attrs(attrs_by_cat, k=k)
print(top_attrs_by_cat.keys())
return top_attrs_by_cat
def generate_attrs(corpus_by_cat, k, save_name):
attrs = dict()
for n in range(1, 3):
ngram_range = (n, n)
top_attrs_by_cat = generate_ngram_attrs(corpus_by_cat, ngram_range, k, attrs)
if save_name is not None:
save_path = Path(ATTR_DIR) / f'{save_name}_{n}-gram.yaml'
with open(save_path, 'w') as f:
yaml.dump(top_attrs_by_cat, f, default_flow_style=False)
print(f'Saved: {save_path}')
save_path = Path(ATTR_DIR) / f'{save_name}_attrs_unfiltered.json'
with open(save_path, 'w') as f:
json.dump(attrs, f)
print(f'Saved: {save_path}')
if __name__ == '__main__':
''
products = load_products(num=40000)
corpus_by_cat_broad = get_corpus(products, category_type='category')
generate_attrs(corpus_by_cat_broad, k=5, save_name='broad')
corpus_by_cat_narrow = get_corpus(products, category_type='query')
generate_attrs(corpus_by_cat_narrow, k=5, save_name='narrow')
# File: WebShop-master/web_agent_site/engine/engine.py
""""""
import os
import re
import json
import random
from collections import defaultdict
from ast import literal_eval
from decimal import Decimal
import cleantext
from tqdm import tqdm
from rank_bm25 import BM25Okapi
from flask import render_template_string
from rich import print
from pyserini.search.lucene import LuceneSearcher
from web_agent_site.utils import BASE_DIR, DEFAULT_FILE_PATH, DEFAULT_REVIEW_PATH, DEFAULT_ATTR_PATH, HUMAN_ATTR_PATH
TEMPLATE_DIR = os.path.join(BASE_DIR, 'templates')
SEARCH_RETURN_N = 50
PRODUCT_WINDOW = 10
TOP_K_ATTR = 10
END_BUTTON = 'Buy Now'
NEXT_PAGE = 'Next >'
PREV_PAGE = '< Prev'
BACK_TO_SEARCH = 'Back to Search'
ACTION_TO_TEMPLATE = {'Description': 'description_page.html', 'Features': 'features_page.html', 'Reviews': 'review_page.html', 'Attributes': 'attributes_page.html'}
def map_action_to_html(action, **kwargs):
(action_name, action_arg) = parse_action(action)
if action_name == 'start':
path = os.path.join(TEMPLATE_DIR, 'search_page.html')
html = render_template_string(read_html_template(path=path), session_id=kwargs['session_id'], instruction_text=kwargs['instruction_text'])
elif action_name == 'search':
path = os.path.join(TEMPLATE_DIR, 'results_page.html')
html = render_template_string(read_html_template(path=path), session_id=kwargs['session_id'], products=kwargs['products'], keywords=kwargs['keywords'], page=kwargs['page'], total=kwargs['total'], instruction_text=kwargs['instruction_text'])
elif action_name == 'click' and action_arg == END_BUTTON:
path = os.path.join(TEMPLATE_DIR, 'done_page.html')
html = render_template_string(read_html_template(path), session_id=kwargs['session_id'], reward=kwargs['reward'], asin=kwargs['asin'], options=kwargs['options'], reward_info=kwargs.get('reward_info'), goal_attrs=kwargs.get('goal_attrs'), purchased_attrs=kwargs.get('purchased_attrs'), goal=kwargs.get('goal'), mturk_code=kwargs.get('mturk_code'), query=kwargs.get('query'), category=kwargs.get('category'), product_category=kwargs.get('product_category'))
elif action_name == 'click' and action_arg in ACTION_TO_TEMPLATE:
path = os.path.join(TEMPLATE_DIR, ACTION_TO_TEMPLATE[action_arg])
html = render_template_string(read_html_template(path), session_id=kwargs['session_id'], product_info=kwargs['product_info'], keywords=kwargs['keywords'], page=kwargs['page'], asin=kwargs['asin'], options=kwargs['options'], instruction_text=kwargs.get('instruction_text'))
elif action_name == 'click':
path = os.path.join(TEMPLATE_DIR, 'item_page.html')
html = render_template_string(read_html_template(path), session_id=kwargs['session_id'], product_info=kwargs['product_info'], keywords=kwargs['keywords'], page=kwargs['page'], asin=kwargs['asin'], options=kwargs['options'], instruction_text=kwargs.get('instruction_text'), show_attrs=kwargs['show_attrs'])
else:
raise ValueError('Action name not recognized.')
return html
def read_html_template(path):
with open(path) as f:
template = f.read()
return template
def parse_action(action):
pattern = re.compile('(.+)\\[(.+)\\]')
m = re.match(pattern, action)
if m is None:
action_name = action
action_arg = None
else:
(action_name, action_arg) = m.groups()
return (action_name, action_arg)
def convert_web_app_string_to_var(name, string):
if name == 'keywords':
keywords = string
if keywords.startswith('['):
keywords = literal_eval(keywords)
else:
keywords = [keywords]
var = keywords
elif name == 'page':
page = string
page = int(page)
var = page
else:
raise ValueError('Name of variable not recognized.')
return var
def get_top_n_product_from_keywords(keywords, search_engine, all_products, product_item_dict, attribute_to_asins=None):
if keywords[0] == '<r>':
top_n_products = random.sample(all_products, k=SEARCH_RETURN_N)
elif keywords[0] == '<a>':
attribute = ' '.join(keywords[1:]).strip()
asins = attribute_to_asins[attribute]
top_n_products = [p for p in all_products if p['asin'] in asins]
elif keywords[0] == '<c>':
category = keywords[1].strip()
top_n_products = [p for p in all_products if p['category'] == category]
elif keywords[0] == '<q>':
query = ' '.join(keywords[1:]).strip()
top_n_products = [p for p in all_products if p['query'] == query]
else:
keywords = ' '.join(keywords)
hits = search_engine.search(keywords, k=SEARCH_RETURN_N)
docs = [search_engine.doc(hit.docid) for hit in hits]
top_n_asins = [json.loads(doc.raw())['id'] for doc in docs]
top_n_products = [product_item_dict[asin] for asin in top_n_asins if asin in product_item_dict]
return top_n_products
def get_product_per_page(top_n_products, page):
return top_n_products[(page - 1) * PRODUCT_WINDOW:page * PRODUCT_WINDOW]
def generate_product_prices(all_products):
product_prices = dict()
for product in all_products:
asin = product['asin']
pricing = product['pricing']
if not pricing:
price = 100.0
elif len(pricing) == 1:
price = pricing[0]
else:
price = random.uniform(*pricing[:2])
product_prices[asin] = price
return product_prices
def init_search_engine(num_products=None):
if num_products == 100:
indexes = 'indexes_100'
elif num_products == 1000:
indexes = 'indexes_1k'
elif num_products == 100000:
indexes = 'indexes_100k'
elif num_products is None:
indexes = 'indexes'
else:
raise NotImplementedError(f'num_products being {num_products} is not supported yet.')
search_engine = LuceneSearcher(os.path.join(BASE_DIR, f'../search_engine/{indexes}'))
return search_engine
def clean_product_keys(products):
for product in products:
product.pop('product_information', None)
product.pop('brand', None)
product.pop('brand_url', None)
product.pop('list_price', None)
product.pop('availability_quantity', None)
product.pop('availability_status', None)
product.pop('total_reviews', None)
product.pop('total_answered_questions', None)
product.pop('seller_id', None)
product.pop('seller_name', None)
product.pop('fulfilled_by_amazon', None)
product.pop('fast_track_message', None)
product.pop('aplus_present', None)
product.pop('small_description_old', None)
print('Keys cleaned.')
return products
def load_products(filepath, num_products=None, human_goals=True):
with open(filepath) as f:
products = json.load(f)
print('Products loaded.')
products = clean_product_keys(products)
all_reviews = dict()
all_ratings = dict()
if human_goals:
with open(HUMAN_ATTR_PATH) as f:
human_attributes = json.load(f)
with open(DEFAULT_ATTR_PATH) as f:
attributes = json.load(f)
with open(HUMAN_ATTR_PATH) as f:
human_attributes = json.load(f)
print('Attributes loaded.')
asins = set()
all_products = []
attribute_to_asins = defaultdict(set)
if num_products is not None:
products = products[:num_products]
for (i, p) in tqdm(enumerate(products), total=len(products)):
asin = p['asin']
if asin == 'nan' or len(asin) > 10:
continue
if asin in asins:
continue
else:
asins.add(asin)
products[i]['category'] = p['category']
products[i]['query'] = p['query']
products[i]['product_category'] = p['product_category']
products[i]['Title'] = p['name']
products[i]['Description'] = p['full_description']
products[i]['Reviews'] = all_reviews.get(asin, [])
products[i]['Rating'] = all_ratings.get(asin, 'N.A.')
for r in products[i]['Reviews']:
if 'score' not in r:
r['score'] = r.pop('stars')
if 'review' not in r:
r['body'] = ''
else:
r['body'] = r.pop('review')
products[i]['BulletPoints'] = p['small_description'] if isinstance(p['small_description'], list) else [p['small_description']]
pricing = p.get('pricing')
if pricing is None or not pricing:
pricing = [100.0]
price_tag = '$100.0'
else:
pricing = [float(Decimal(re.sub('[^\\d.]', '', price))) for price in pricing.split('$')[1:]]
if len(pricing) == 1:
price_tag = f'${pricing[0]}'
else:
price_tag = f'${pricing[0]} to ${pricing[1]}'
pricing = pricing[:2]
products[i]['pricing'] = pricing
products[i]['Price'] = price_tag
options = dict()
customization_options = p['customization_options']
option_to_image = dict()
if customization_options:
for (option_name, option_contents) in customization_options.items():
if option_contents is None:
continue
option_name = option_name.lower()
option_values = []
for option_content in option_contents:
option_value = option_content['value'].strip().replace('/', ' | ').lower()
option_image = option_content.get('image', None)
option_values.append(option_value)
option_to_image[option_value] = option_image
options[option_name] = option_values
products[i]['options'] = options
products[i]['option_to_image'] = option_to_image
if asin in attributes and 'attributes' in attributes[asin]:
products[i]['Attributes'] = attributes[asin]['attributes']
else:
products[i]['Attributes'] = ['DUMMY_ATTR']
if human_goals:
if asin in human_attributes:
products[i]['instructions'] = human_attributes[asin]
else:
products[i]['instruction_text'] = attributes[asin].get('instruction', None)
products[i]['instruction_attributes'] = attributes[asin].get('instruction_attributes', None)
products[i]['MainImage'] = p['images'][0]
products[i]['query'] = p['query'].lower().strip()
all_products.append(products[i])
for p in all_products:
for a in p['Attributes']:
attribute_to_asins[a].add(p['asin'])
product_item_dict = {p['asin']: p for p in all_products}
product_prices = generate_product_prices(all_products)
return (all_products, product_item_dict, product_prices, attribute_to_asins)
# File: WebShop-master/web_agent_site/engine/goal.py
""""""
import itertools
import random
import spacy
from collections import defaultdict
from rich import print
from thefuzz import fuzz
from web_agent_site.engine.normalize import normalize_color
nlp = spacy.load('en_core_web_sm')
PRICE_RANGE = [10.0 * i for i in range(1, 100)]
def get_goals(all_products, product_prices, human_goals=True):
if human_goals:
return get_human_goals(all_products, product_prices)
else:
return get_synthetic_goals(all_products, product_prices)
def get_human_goals(all_products, product_prices):
goals = []
cnt_atts = defaultdict(int)
cnt = 0
for item in all_products:
asin = item['asin']
if 'instructions' not in item:
continue
for product in item['instructions']:
attributes = product['instruction_attributes']
if len(attributes) == 0:
cnt += 1
continue
if product_prices is not None:
price = product_prices[asin]
price_range = [p for p in PRICE_RANGE if p > price][:4]
if len(price_range) >= 2:
(_, price_upper) = sorted(random.sample(price_range, 2))
price_text = f', and price lower than {price_upper:.2f} dollars'
else:
price_upper = 1000000
price_text = ''
else:
price_upper = 1000000
goals.append({'asin': asin, 'category': item['category'], 'query': item['query'], 'name': item['name'], 'product_category': item['product_category'], 'instruction_text': product['instruction'].strip('.') + price_text, 'attributes': attributes, 'price_upper': price_upper, 'goal_options': product['instruction_options']})
for att in attributes:
cnt_atts[att] += 1
for goal in goals:
goal['weight'] = 1
print(cnt, 'skipped')
return goals
def get_synthetic_goals(all_products, product_prices):
goals = []
cnt_atts = defaultdict(int)
for product in all_products:
if 'instruction_text' not in product or product['instruction_text'] is None:
continue
product_goals = []
asin = product['asin']
attributes = product['instruction_attributes']
assert len(attributes) > 0
if product_prices is not None:
price = product_prices[asin]
price_range = [p for p in PRICE_RANGE if p > price][:4]
if len(price_range) >= 2:
(_, price_upper) = sorted(random.sample(price_range, 2))
price_text = f', and price lower than {price_upper:.2f} dollars'
else:
price_upper = 1000000
price_text = ''
else:
price_upper = 1000000
price_text = ''
instruction_text = product['instruction_text']
options = product['options']
option_names = sorted(options)
combinations = list(itertools.product(*(options[option_name] for option_name in option_names)))
for combination in combinations:
goal_options = dict()
for (i, o) in enumerate(combination):
goal_options[option_names[i]] = o
option_text = ', and '.join([f'{k}: {v}' for (k, v) in goal_options.items()])
option_text = ' with ' + option_text if option_text else ''
product_goals.append({'asin': asin, 'category': product['category'], 'query': product['query'], 'name': product['name'], 'product_category': product['product_category'], 'instruction_text': f'{instruction_text}{option_text}{price_text}', 'attributes': attributes, 'price_upper': price_upper, 'goal_options': goal_options, 'name': product['Title']})
for att in attributes:
cnt_atts[att] += 1
goals += product_goals
for goal in goals:
goal['weight'] = sum((1.0 / cnt_atts[att] for att in goal['attributes'])) / len(goal['attributes'])
return goals
def get_type_reward(purchased_product, goal):
query_match = purchased_product['query'] == goal['query']
purchased_product_category = [x.strip() for x in purchased_product['product_category'].split('›')]
goal_product_category = [x.strip() for x in goal['product_category'].split('›')]
category_match = len(set(purchased_product_category) & set(goal_product_category)) >= 2
purchased_type = purchased_product['name']
desired_type = goal['name']
purchased_type_parse = nlp(purchased_type)
desired_type_parse = nlp(desired_type)
purchased_type_parse = [t.text.lower() for t in purchased_type_parse if t.pos_ in ('PNOUN', 'NOUN', 'PROPN')]
desired_type_parse = [t.text.lower() for t in desired_type_parse if t.pos_ in ('PNOUN', 'NOUN', 'PROPN')]
n_intersect_type = len(set(purchased_type_parse) & set(desired_type_parse))
if len(desired_type_parse) == 0:
title_score = 0.2
else:
title_score = n_intersect_type / len(desired_type_parse)
r_type = 1.0
match = query_match or category_match or title_score > 0.2
if not match:
r_type = 0.5
if title_score < 0.1:
r_type = 0.1
if title_score == 0.0:
r_type = 0.0
return dict(r_type=r_type, query_match=query_match, category_match=category_match, title_score=title_score)
def get_attribute_reward(purchased_product, goal):
purchased_attrs = purchased_product['Attributes']
goal_attrs = goal['attributes']
num_attr_matches = 0
for g_attr in goal_attrs:
matched = False
for p_attr in purchased_attrs:
score = fuzz.token_set_ratio(p_attr, g_attr)
if score > 85:
num_attr_matches += 1
matched = True
break
if not matched and (g_attr in purchased_product['Title'].lower() or g_attr in ' '.join(purchased_product['BulletPoints']).lower() or g_attr in purchased_product['Description'].lower()):
num_attr_matches += 1
matched = True
r_attr = num_attr_matches / len(goal_attrs)
return (r_attr, num_attr_matches)
def get_option_reward(purchased_options, goal_options):
purchased_options = [normalize_color(o) for o in purchased_options]
goal_options = [normalize_color(o) for o in goal_options]
num_option_matches = 0
for g_option in goal_options:
for p_option in purchased_options:
score = fuzz.token_set_ratio(p_option, g_option)
if score > 85:
num_option_matches += 1
break
r_option = num_option_matches / len(goal_options) if len(goal_options) > 0 else None
return (r_option, num_option_matches)
def get_reward(purchased_product, goal, price, options, **kwargs):
r_type_dict = get_type_reward(purchased_product, goal)
r_price = price <= goal['price_upper'] if goal['price_upper'] > 0 else None
(r_att, num_attr_matches) = get_attribute_reward(purchased_product, goal)
(r_option, num_option_matches) = get_option_reward(list(options.values()), goal['goal_options'].items() if isinstance(goal['goal_options'], dict) else goal['goal_options'])
total_reward = (num_attr_matches + num_option_matches + r_price) / (len(goal['attributes']) + len(goal['goal_options']) + 1)
total_reward *= r_type_dict['r_type']
if kwargs.get('verbose', False):
info = {'r_type': r_type_dict['r_type'], 'r_att': r_att, 'w_att': len(goal['attributes']) / (len(goal['attributes']) + len(goal['goal_options']) + 1), 'query_match': r_type_dict['query_match'], 'category_match': r_type_dict['category_match'], 'title_score': r_type_dict['title_score']}
if r_option is not None:
info['r_option'] = r_option
info['w_option'] = len(goal['goal_options']) / (len(goal['attributes']) + len(goal['goal_options']) + 1)
if r_price is not None:
info['r_price'] = r_price
info['w_price'] = 1 / (len(goal['attributes']) + len(goal['goal_options']) + 1)
return (total_reward, info)
return total_reward
# File: WebShop-master/web_agent_site/engine/normalize.py
import re
from typing import Tuple
COLOR_SET = ['alabaster', 'apricot', 'aqua', 'ash', 'asphalt', 'azure', 'banana', 'beige', 'black', 'blue', 'blush', 'bordeaux', 'bronze', 'brown', 'burgundy', 'camel', 'camo', 'caramel', 'champagne', 'charcoal', 'cheetah', 'chestnut', 'chocolate', 'christmas', 'coffee', 'cognac', 'copper', 'coral', 'cranberry', 'cream', 'crystal', 'dark', 'denim', 'eggplant', 'elephant', 'espresso', 'fuchsia', 'gold', 'granite', 'grape', 'graphite', 'grass', 'gray', 'green', 'grey', 'heather', 'indigo', 'ivory', 'ivy', 'khaki', 'lavender', 'lemon', 'leopard', 'light', 'lilac', 'lime', 'magenta', 'maroon', 'mauve', 'merlot', 'midnight', 'mint', 'mocha', 'multicolor', 'mushroom', 'mustard', 'natural', 'navy', 'nude', 'olive', 'orange', 'peach', 'pewter', 'pink', 'plum', 'purple', 'rainbow', 'red', 'rose', 'royal', 'rust', 'sand', 'sapphire', 'seashell', 'silver', 'skull', 'slate', 'steel', 'stone', 'stonewash', 'sunflower', 'tan', 'taupe', 'teal', 'tiger', 'turquoise', 'violet', 'walnut', 'wheat', 'white', 'wine', 'yellow']
SIZE_SET = ['xx-large', '3x-large', '4x-large', '5x-large', 'x-large', 'x-small', 'medium', 'large', 'small', 'queen', 'twin', 'full', 'king', 'one size', 'pack']
SIZE_PATTERNS = [re.compile('(.*)neck(.*)sleeve'), re.compile('(.*) women \\| (.*) men'), re.compile('(.*)w x(.*)l'), re.compile('(.*)w by (.*)l'), re.compile('(.*)w x(.*)h'), re.compile('(.*)wide'), re.compile('(.*)x-wide'), re.compile('(.*)narrow'), re.compile('(.*)petite'), re.compile('(.*)inch'), re.compile('(.*)plus'), re.compile('(.*)mm'), re.compile('women(.*)'), re.compile('(.*)x(.*)'), re.compile('(.*)ft'), re.compile('(.*)feet'), re.compile('(.*)meter'), re.compile('(.*)yards'), re.compile('(.*)\\*(.*)'), re.compile('(.*)\\-(.*)'), re.compile('(\\d+)"$'), re.compile('(\\d+)f$'), re.compile('(\\d+)m$'), re.compile('(\\d+)cm$'), re.compile('(\\d+)g$')]
SIZE_PATTERNS = [re.compile(s) for s in SIZE_SET] + SIZE_PATTERNS
def normalize_color(color_string: str) -> str:
for norm_color in COLOR_SET:
if norm_color in color_string:
return norm_color
return color_string
def normalize_color_size(product_prices: dict) -> Tuple[dict, dict]:
(all_colors, all_sizes) = (set(), set())
for ((_, color, size), _) in product_prices.items():
all_colors.add(color.lower())
all_sizes.add(size.lower())
color_mapping = {'N.A.': 'not_matched'}
for c in all_colors:
matched = False
for base in COLOR_SET:
if base in c:
color_mapping[c] = base
matched = True
break
if not matched:
color_mapping[c] = 'not_matched'
size_mapping = {'N.A.': 'not_matched'}
for s in all_sizes:
matched = False
for pattern in SIZE_PATTERNS:
m = re.search(pattern, s)
if m is not None:
matched = True
size_mapping[s] = pattern.pattern
break
if not matched:
if s.replace('.', '', 1).isdigit():
size_mapping[s] = 'numeric_size'
matched = True
if not matched:
size_mapping[s] = 'not_matched'
return (color_mapping, size_mapping)
# File: WebShop-master/web_agent_site/envs/__init__.py
from gym.envs.registration import register
from web_agent_site.envs.web_agent_site_env import WebAgentSiteEnv
from web_agent_site.envs.web_agent_text_env import WebAgentTextEnv
register(id='WebAgentSiteEnv-v0', entry_point='web_agent_site.envs:WebAgentSiteEnv')
register(id='WebAgentTextEnv-v0', entry_point='web_agent_site.envs:WebAgentTextEnv')
# File: WebShop-master/web_agent_site/envs/web_agent_site_env.py
import gym
import random
import requests
import string
import time
from bs4 import BeautifulSoup
from bs4.element import Comment
from gym import spaces
from os.path import join, dirname, abspath
from selenium import webdriver
from selenium.webdriver.chrome.service import Service
from selenium.webdriver.chrome.options import Options
from selenium.webdriver.common.keys import Keys
from selenium.common.exceptions import ElementNotInteractableException
from web_agent_site.engine.engine import parse_action, END_BUTTON
class WebAgentSiteEnv(gym.Env):
def __init__(self, observation_mode='html', **kwargs):
super(WebAgentSiteEnv, self).__init__()
self.observation_mode = observation_mode
self.kwargs = kwargs
service = Service(join(dirname(abspath(__file__)), 'chromedriver'))
options = Options()
if 'render' not in kwargs or not kwargs['render']:
options.add_argument('--headless')
self.browser = webdriver.Chrome(service=service, options=options)
self.text_to_clickable = None
self.assigned_session = kwargs.get('session')
self.session = None
self.reset()
def step(self, action):
reward = 0.0
done = False
info = None
(action_name, action_arg) = parse_action(action)
if action_name == 'search':
try:
search_bar = self.browser.find_element_by_id('search_input')
except Exception:
pass
else:
search_bar.send_keys(action_arg)
search_bar.submit()
elif action_name == 'click':
try:
self.text_to_clickable[action_arg].click()
except ElementNotInteractableException:
button = self.text_to_clickable[action_arg]
self.browser.execute_script('arguments[0].click();', button)
reward = self.get_reward()
if action_arg == END_BUTTON:
done = True
elif action_name == 'end':
done = True
else:
print('Invalid action. No action performed.')
if 'pause' in self.kwargs:
time.sleep(self.kwargs['pause'])
return (self.observation, reward, done, info)
def get_available_actions(self):
try:
search_bar = self.browser.find_element_by_id('search_input')
except Exception:
has_search_bar = False
else:
has_search_bar = True
buttons = self.browser.find_elements_by_class_name('btn')
product_links = self.browser.find_elements_by_class_name('product-link')
buying_options = self.browser.find_elements_by_css_selector("input[type='radio']")
self.text_to_clickable = {f'{b.text}': b for b in buttons + product_links}
for opt in buying_options:
opt_value = opt.get_attribute('value')
self.text_to_clickable[f'{opt_value}'] = opt
return dict(has_search_bar=has_search_bar, clickables=list(self.text_to_clickable.keys()))
def _parse_html(self, html=None, url=None):
if html is None:
if url is not None:
html = requests.get(url)
else:
html = self.state['html']
html_obj = BeautifulSoup(html, 'html.parser')
return html_obj
def get_reward(self):
html_obj = self._parse_html()
r = html_obj.find(id='reward')
r = float(r.findChildren('pre')[0].string) if r is not None else 0.0
return r
def get_instruction_text(self):
html_obj = self._parse_html(self.browser.page_source)
instruction_text = html_obj.find(id='instruction-text').h4.text
return instruction_text
def convert_html_to_text(self, html):
texts = self._parse_html(html).findAll(text=True)
visible_texts = filter(tag_visible, texts)
observation = ' [SEP] '.join((t.strip() for t in visible_texts if t != '\n'))
return observation
@property
def state(self):
return dict(url=self.browser.current_url, html=self.browser.page_source, instruction_text=self.instruction_text)
@property
def observation(self):
html = self.state['html']
if self.observation_mode == 'html':
return html
elif self.observation_mode == 'text':
return self.convert_html_to_text(html)
else:
raise ValueError(f'Observation mode {self.observation_mode} not supported.')
@property
def action_space(self):
return NotImplementedError
@property
def observation_space(self):
return NotImplementedError
def reset(self):
if self.assigned_session is not None:
self.session = self.assigned_session
else:
self.session = ''.join(random.choices(string.ascii_lowercase, k=5))
init_url = f'http://127.0.0.1:3000/{self.session}'
self.browser.get(init_url)
self.instruction_text = self.get_instruction_text()
return (self.observation, None)
def render(self, mode='human'):
return NotImplementedError
def close(self):
self.browser.close()
print('Browser closed.')
def tag_visible(element):
ignore = {'style', 'script', 'head', 'title', 'meta', '[document]'}
return element.parent.name not in ignore and (not isinstance(element, Comment))
# File: WebShop-master/web_agent_site/envs/web_agent_text_env.py
import gym
import json
import random
import string
import time
import torch
import numpy as np
from bs4 import BeautifulSoup
from bs4.element import Comment
from collections import defaultdict
from flask import Flask
from web_agent_site.engine.engine import load_products, init_search_engine, get_top_n_product_from_keywords, map_action_to_html, parse_action, get_product_per_page, ACTION_TO_TEMPLATE, END_BUTTON, NEXT_PAGE, PREV_PAGE, BACK_TO_SEARCH
from web_agent_site.engine.goal import get_reward, get_goals
from web_agent_site.utils import DEFAULT_FILE_PATH, FEAT_CONV, FEAT_IDS, random_idx
app = Flask(__name__)
class WebAgentTextEnv(gym.Env):
def __init__(self, observation_mode='html', file_path=DEFAULT_FILE_PATH, server=None, **kwargs):
super(WebAgentTextEnv, self).__init__()
self.observation_mode = observation_mode
self.kwargs = kwargs
self.file_path = file_path
self.base_url = 'http://127.0.0.1:3000'
self.server = SimServer(self.base_url, self.file_path, self.kwargs.get('filter_goals'), self.kwargs.get('limit_goals', -1), self.kwargs.get('num_products'), self.kwargs.get('human_goals'), self.kwargs.get('show_attrs', False)) if server is None else server
self.browser = SimBrowser(self.server)
self.session = self.kwargs.get('session')
self.session_prefix = self.kwargs.get('session_prefix')
if self.kwargs.get('get_image', 0):
self.feats = torch.load(FEAT_CONV)
self.ids = torch.load(FEAT_IDS)
self.ids = {url: idx for (idx, url) in enumerate(self.ids)}
self.prev_obs = []
self.prev_actions = []
self.num_prev_obs = self.kwargs.get('num_prev_obs', 0)
self.num_prev_actions = self.kwargs.get('num_prev_actions', 0)
self.reset()
def step(self, action):
info = None
self.get_available_actions()
(action_name, action_arg) = parse_action(action)
if action_arg is not None:
action_arg = action_arg.lower()
if action_name == 'search' and action_arg is not None and (action_arg != ''):
status = self.browser.search(action_arg)
elif action_name == 'click' and action_arg in self.text_to_clickable.keys() and (action_arg != 'search'):
status = self.browser.click(action_arg, self.text_to_clickable)
else:
status = dict(reward=0, done=False)
ob = self.observation
text_list = [ob]
self.prev_actions.append(action)
for i in range(1, 1 + max(self.num_prev_obs, self.num_prev_actions)):
if len(self.prev_actions) >= i and self.num_prev_actions >= i:
text_list.append(self.prev_actions[-i])
if len(self.prev_obs) >= i and self.num_prev_obs >= i:
text_list.append(self.prev_obs[-i])
state = ' [SEP] '.join(text_list[::-1])
self.prev_obs.append(ob)
return (state, status['reward'], status['done'], info)
def get_available_actions(self):
html_obj = self._parse_html()
search_bar = html_obj.find(id='search_input')
has_search_bar = True if search_bar is not None else False
buttons = html_obj.find_all(class_='btn')
product_links = html_obj.find_all(class_='product-link')
buying_options = html_obj.select('input[type="radio"]')
self.text_to_clickable = {f'{b.get_text()}'.lower(): b for b in buttons + product_links}
for opt in buying_options:
opt_value = opt.get('value')
self.text_to_clickable[f'{opt_value}'] = opt
return dict(has_search_bar=has_search_bar, clickables=list(self.text_to_clickable.keys()))
def get_image(self):
html_obj = self._parse_html(self.browser.page_source)
image_url = html_obj.find(id='product-image')
if image_url is not None:
image_url = image_url['src']
if image_url in self.ids:
image_idx = self.ids[image_url]
image = self.feats[image_idx]
return image
return torch.zeros(512)
def get_instruction_text(self):
html_obj = self._parse_html(self.browser.page_source)
instruction_text = html_obj.find(id='instruction-text').h4.text
return instruction_text
def _parse_html(self, html=None):
if html is None:
html = self.state['html']
html_obj = BeautifulSoup(html, 'html.parser')
return html_obj
@property
def observation(self):
html = self.state['html']
if self.observation_mode == 'html':
return html
elif self.observation_mode == 'text':
return self.convert_html_to_text(html, simple=True)
elif self.observation_mode == 'text_rich':
return self.convert_html_to_text(html, simple=False)
elif self.observation_mode == 'url':
return self.state['url']
else:
raise ValueError(f'Observation mode {self.observation_mode} not supported.')
@property
def state(self):
return dict(url=self.browser.current_url, html=self.browser.page_source, instruction_text=self.instruction_text)
def convert_html_to_text(self, html, simple=False):
texts = self._parse_html(html).findAll(text=True)
visible_texts = filter(tag_visible, texts)
if simple:
return ' [SEP] '.join((t.strip() for t in visible_texts if t != '\n'))
else:
observation = ''
for t in visible_texts:
if t == '\n':
continue
if t.parent.name == 'button':
processed_t = f'[button] {t} [button_]'
elif t.parent.name == 'label':
if f'"{t}"' in self.state['url']:
processed_t = f' [clicked button] {t} [clicked button_]'
observation = f'You have clicked {t}.\n' + observation
else:
processed_t = f' [button] {t} [button_]'
elif t.parent.get('class') == ['product-link']:
if f'{t}' in self.server.user_sessions[self.session]['asins']:
processed_t = f'\n[clicked button] {t} [clicked button_]'
else:
processed_t = f'\n[button] {t} [button_]'
else:
processed_t = str(t)
observation += processed_t + '\n'
return observation
def reset(self, session=None, instruction_text=None):
session_int = None
if session is not None:
self.session = str(session)
if isinstance(session, int):
session_int = session
else:
self.session = ''.join(random.choices(string.ascii_lowercase, k=10))
if self.session_prefix is not None:
self.session = self.session_prefix + self.session
init_url = f'{self.base_url}/{self.session}'
self.browser.get(init_url, session_id=self.session, session_int=session_int)
self.text_to_clickable = None
self.instruction_text = self.get_instruction_text() if instruction_text is None else instruction_text
obs = self.observation
self.prev_obs = [obs]
self.prev_actions = []
return (obs, None)
def render(self, mode='human'):
pass
def close(self):
pass
def tag_visible(element):
ignore = {'style', 'script', 'head', 'title', 'meta', '[document]'}
return element.parent.name not in ignore and (not isinstance(element, Comment))
class SimServer:
def __init__(self, base_url, file_path, filter_goals=None, limit_goals=-1, num_products=None, human_goals=0, show_attrs=False):
self.base_url = base_url
(self.all_products, self.product_item_dict, self.product_prices, _) = load_products(filepath=file_path, num_products=num_products, human_goals=human_goals)
self.search_engine = init_search_engine(num_products=num_products)
self.goals = get_goals(self.all_products, self.product_prices, human_goals)
self.show_attrs = show_attrs
random.seed(233)
random.shuffle(self.goals)
if filter_goals is not None:
self.goals = [goal for (i, goal) in enumerate(self.goals) if filter_goals(i, goal)]
if limit_goals != -1 and limit_goals < len(self.goals):
self.weights = [goal['weight'] for goal in self.goals]
self.cum_weights = [0] + np.cumsum(self.weights).tolist()
idxs = []
while len(idxs) < limit_goals:
idx = random_idx(self.cum_weights)
if idx not in idxs:
idxs.append(idx)
self.goals = [self.goals[i] for i in idxs]
print(f'Loaded {len(self.goals)} goals.')
self.weights = [goal['weight'] for goal in self.goals]
self.cum_weights = [0] + np.cumsum(self.weights).tolist()
self.user_sessions = dict()
self.search_time = 0
self.render_time = 0
self.sample_time = 0
self.assigned_instruction_text = None
@app.route('/', methods=['GET', 'POST'])
def index(self, session_id, **kwargs):
html = map_action_to_html('start', session_id=session_id, instruction_text=kwargs['instruction_text'])
url = f'{self.base_url}/{session_id}'
return (html, url)
@app.route('/', methods=['GET', 'POST'])
def search_results(self, session_id, **kwargs):
session = self.user_sessions[session_id]
keywords = kwargs['keywords']
assert isinstance(keywords, list)
page = 1 if 'page' not in kwargs else kwargs['page']
session['page'] = page
session['keywords'] = keywords
session['actions']['search'] += 1
session['asin'] = None
session['options'] = {}
old_time = time.time()
top_n_products = get_top_n_product_from_keywords(keywords, self.search_engine, self.all_products, self.product_item_dict)
self.search_time += time.time() - old_time
products = get_product_per_page(top_n_products, page)
keywords_url_string = '+'.join(keywords)
url = f'{self.base_url}/search_results/{session_id}/{keywords_url_string}/{page}'
old_time = time.time()
html = map_action_to_html('search', session_id=session_id, products=products, keywords=session['keywords'], page=page, total=len(top_n_products), instruction_text=session['goal']['instruction_text'])
self.render_time += time.time() - old_time
return (html, url)
@app.route('/', methods=['GET', 'POST'])
def item_page(self, session_id, **kwargs):
session = self.user_sessions[session_id]
clickable_name = kwargs['clickable_name']
text_to_clickable = kwargs['text_to_clickable']
clickable = text_to_clickable[clickable_name]
if clickable.get('class') is not None and clickable.get('class')[0] == 'product-link':
session['asin'] = clickable_name.upper()
session['actions']['asin'] += 1
session['asins'].add(session['asin'])
elif clickable.get('name') is not None:
clickable_key = clickable['name'].lower()
session['options'][clickable_key] = clickable_name
session['actions']['options'] += 1
product_info = self.product_item_dict[session['asin']]
keywords_url_string = '+'.join(session['keywords'])
option_string = json.dumps(session['options'])
url = f"{self.base_url}/item_page/{session_id}/{session['asin']}/{keywords_url_string}/{session['page']}/{option_string}"
html = map_action_to_html('click', session_id=session_id, product_info=product_info, keywords=session['keywords'], page=session['page'], asin=session['asin'], options=session['options'], instruction_text=session['goal']['instruction_text'], show_attrs=self.show_attrs)
return (html, url)
@app.route('/', methods=['GET', 'POST'])
def item_sub_page(self, session_id, **kwargs):
session = self.user_sessions[session_id]
clickable_name = kwargs['clickable_name']
for k in ACTION_TO_TEMPLATE:
if clickable_name.lower() == k.lower():
clickable_name = k
break
product_info = self.product_item_dict[session['asin']]
session['actions'][clickable_name] += 1
keywords_url_string = '+'.join(session['keywords'])
url = f"{self.base_url}/item_sub_page/{session_id}/{session['asin']}/{keywords_url_string}/{session['page']}/{clickable_name}/{session['options']}"
html = map_action_to_html(f'click[{clickable_name}]', session_id=session_id, product_info=product_info, keywords=session['keywords'], page=session['page'], asin=session['asin'], options=session['options'], instruction_text=session['goal']['instruction_text'])
return (html, url)
@app.route('/', methods=['GET', 'POST'])
def done(self, session_id, **kwargs):
session = self.user_sessions[session_id]
goal = self.user_sessions[session_id]['goal']
purchased_product = self.product_item_dict[session['asin']]
session['actions']['purchase'] += 1
price = self.product_prices.get(session['asin'])
(reward, info) = get_reward(purchased_product, goal, price=price, options=session['options'], verbose=True)
self.user_sessions[session_id]['verbose_info'] = info
self.user_sessions[session_id]['done'] = True
self.user_sessions[session_id]['reward'] = reward
url = f"{self.base_url}/done/{session_id}/{session['asin']}/{session['options']}"
html = map_action_to_html(f'click[{END_BUTTON}]', session_id=session_id, reward=reward, asin=session['asin'], options=session['options'], instruction_text=session['goal']['instruction_text'])
return (html, url, reward)
def receive(self, session_id, current_url, session_int=None, **kwargs):
status = dict(reward=0.0, done=False)
with app.app_context(), app.test_request_context():
if session_id not in self.user_sessions:
idx = session_int if session_int is not None and isinstance(session_int, int) else random_idx(self.cum_weights)
goal = self.goals[idx]
instruction_text = goal['instruction_text']
self.user_sessions[session_id] = {'goal': goal, 'done': False}
else:
instruction_text = self.user_sessions[session_id]['goal']['instruction_text']
if self.assigned_instruction_text is not None:
instruction_text = self.assigned_instruction_text
self.user_sessions[session_id]['goal']['instruction_text'] = instruction_text
session = self.user_sessions[session_id]
if not kwargs:
kwargs['instruction_text'] = instruction_text
(html, url) = self.index(session_id, **kwargs)
self.user_sessions[session_id].update({'keywords': None, 'page': None, 'asin': None, 'asins': set(), 'options': dict(), 'actions': defaultdict(int)})
elif 'keywords' in kwargs:
(html, url) = self.search_results(session_id, **kwargs)
elif 'clickable_name' in kwargs:
clickable_name = kwargs['clickable_name'].lower()
if clickable_name == END_BUTTON.lower():
(html, url, reward) = self.done(session_id, **kwargs)
status['reward'] = reward
status['done'] = True
elif clickable_name == BACK_TO_SEARCH.lower():
(html, url, status) = self.receive(session_id, current_url)
elif clickable_name == NEXT_PAGE.lower() and self.get_page_name(current_url) == 'search_results':
(html, url, status) = self.receive(session_id, current_url, keywords=session['keywords'], page=session['page'] + 1)
elif clickable_name == PREV_PAGE.lower() and self.get_page_name(current_url) == 'search_results':
(html, url, status) = self.receive(session_id, current_url, keywords=session['keywords'], page=session['page'] - 1)
elif clickable_name == PREV_PAGE.lower() and self.get_page_name(current_url) == 'item_sub_page':
(html, url) = self.item_page(session_id, **kwargs)
elif clickable_name == PREV_PAGE.lower() and self.get_page_name(current_url) == 'item_page':
(html, url) = self.search_results(session_id, keywords=session['keywords'], page=session['page'], **kwargs)
elif clickable_name in [k.lower() for k in ACTION_TO_TEMPLATE]:
(html, url) = self.item_sub_page(session_id, **kwargs)
else:
(html, url) = self.item_page(session_id, **kwargs)
return (html, url, status)
def get_page_name(self, url):
if url is None:
return None
page_names = ['search_results', 'item_page', 'item_sub_page', 'done']
for page_name in page_names:
if page_name in url:
return page_name
return ''
class SimBrowser:
def __init__(self, server):
self.server = server
self.current_url = None
self.page_source = None
self.session_id = None
def get(self, url, session_id=None, session_int=None):
self.session_id = url.split('/')[-1] if session_id is None else session_id
(self.page_source, _, _) = self.server.receive(self.session_id, self.current_url, session_int=session_int)
self.current_url = url
def click(self, clickable_name, text_to_clickable):
(self.page_source, self.current_url, status) = self.server.receive(self.session_id, current_url=self.current_url, clickable_name=clickable_name, text_to_clickable=text_to_clickable)
return status
def search(self, keywords):
if isinstance(keywords, str):
keywords = keywords.split(' ')
(self.page_source, self.current_url, status) = self.server.receive(self.session_id, current_url=self.current_url, keywords=keywords)
return status
# File: WebShop-master/web_agent_site/models/models.py
""""""
import random
random.seed(4)
class BasePolicy:
def __init__(self):
pass
def forward(observation, available_actions):
raise NotImplementedError
class HumanPolicy(BasePolicy):
def __init__(self):
super().__init__()
def forward(self, observation, available_actions):
action = input('> ')
return action
class RandomPolicy(BasePolicy):
def __init__(self):
super().__init__()
def forward(self, observation, available_actions):
if available_actions['has_search_bar']:
action = 'search[shoes]'
else:
action_arg = random.choice(available_actions['clickables'])
action = f'click[{action_arg}]'
return action
# File: WebShop-master/web_agent_site/utils.py
import bisect
import hashlib
import logging
import random
from os.path import dirname, abspath, join
BASE_DIR = dirname(abspath(__file__))
DEBUG_PROD_SIZE = None
DEFAULT_ATTR_PATH = join(BASE_DIR, '../data/items_ins_v2_1000.json')
DEFAULT_FILE_PATH = join(BASE_DIR, '../data/items_shuffle_1000.json')
DEFAULT_REVIEW_PATH = join(BASE_DIR, '../data/reviews.json')
FEAT_CONV = join(BASE_DIR, '../data/feat_conv.pt')
FEAT_IDS = join(BASE_DIR, '../data/feat_ids.pt')
HUMAN_ATTR_PATH = join(BASE_DIR, '../data/items_human_ins.json')
HUMAN_ATTR_PATH = join(BASE_DIR, '../data/items_human_ins.json')
def random_idx(cum_weights):
pos = random.uniform(0, cum_weights[-1])
idx = bisect.bisect(cum_weights, pos)
idx = min(idx, len(cum_weights) - 2)
return idx
def setup_logger(session_id, user_log_dir):
logger = logging.getLogger(session_id)
formatter = logging.Formatter('%(message)s')
file_handler = logging.FileHandler(user_log_dir / f'{session_id}.jsonl', mode='w')
file_handler.setFormatter(formatter)
logger.setLevel(logging.INFO)
logger.addHandler(file_handler)
return logger
def generate_mturk_code(session_id: str) -> str:
sha = hashlib.sha1(session_id.encode())
return sha.hexdigest()[:10].upper()