File size: 18,045 Bytes
e551dda
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
import argparse
import copy
import json
import os
import random
import time

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import scipy
import torch
import torch.nn.functional as F
import torch.optim as optim
from matplotlib import cm
from sklearn.metrics import (auc, explained_variance_score, f1_score,
                             mean_absolute_error, mean_squared_error,
                             precision_score, r2_score, recall_score,
                             roc_auc_score, roc_curve)
from torch.nn.functional import softmax
from torch_geometric.utils import subgraph

torch.autograd.set_detect_anomaly(True)
import math
import pickle
import time
from datetime import date, datetime, timedelta

import torch.nn as nn
import torch_geometric
import torchvision.datasets
import torchvision.models
import torchvision.transforms as transforms
from torch.utils.tensorboard import SummaryWriter
from torch_geometric.nn import GIN, MLP, GATConv
from torch_geometric.nn.pool import global_add_pool, global_mean_pool
from torch_geometric.utils import add_self_loops

import dataset
import model_new
import util
from dataset import label_mapping, reverse_label_mapping
from model_new import Smodel

blue = lambda x: '\033[94m' + x + '\033[0m'
red = lambda x: '\033[31m' + x + '\033[0m'
green = lambda x: '\033[32m' + x + '\033[0m'
yellow = lambda x: '\033[33m' + x + '\033[0m'
greenline = lambda x: '\033[42m' + x + '\033[0m'
yellowline = lambda x: '\033[43m' + x + '\033[0m'

def get_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('--model',default="our", type=str)
    parser.add_argument('--train_batch', default=64, type=int)
    parser.add_argument('--test_batch', default=128, type=int)
    parser.add_argument('--share', type=str, default="0")
    parser.add_argument('--edge_rep', type=str, default="True")
    parser.add_argument('--batchnorm', type=str, default="True")
    parser.add_argument('--extent_norm', type=str, default="T")
    parser.add_argument('--spanning_tree', type=str, default="T")
    
    parser.add_argument('--loss_coef', default=0.1, type=float)
    parser.add_argument('--h_ch', default=512, type=int)
    parser.add_argument('--localdepth', type=int, default=1)
    parser.add_argument('--num_interactions', type=int, default=4)
    parser.add_argument('--finaldepth', type=int, default=4)
    parser.add_argument('--classifier_depth', type=int, default=4)
    parser.add_argument('--dropout', type=float, default=0.0)

    parser.add_argument('--dataset', type=str, default='mnist')
    parser.add_argument('--log', type=str, default="True")   
    parser.add_argument('--test_per_round', type=int, default=10)
    parser.add_argument('--patience', type=int, default=30)  #scheduler
    parser.add_argument('--nepoch', type=int, default=301)
    parser.add_argument('--lr', type=float, default=1e-4)
    parser.add_argument('--manualSeed', type=str, default="False")
    parser.add_argument('--man_seed', type=int, default=12345)
    args = parser.parse_args()
    args.log=True if args.log=="True" else False 
    args.edge_rep=True if args.edge_rep=="True" else False
    args.batchnorm=True if args.batchnorm=="True" else False
    args.save_dir=os.path.join('./save/',args.dataset,args.model)
    args.manualSeed=True if args.manualSeed=="True" else False
    return args

args = get_args()
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
criterion=nn.CrossEntropyLoss()
if args.dataset in ["mnist"]:
    x_out=90
    args.data_dir='data/multi_mnist_with_index.pkl'
elif args.dataset in ["mnist_sparse"]:
    x_out=90
    args.data_dir='data/multi_mnist_sparse.pkl'
elif args.dataset in ["building"]:
    x_out=100
    args.data_dir='data/building_with_index.pkl'
elif args.dataset in ["mbuilding"]:
    x_out=100
    args.data_dir='data/mp_building.pkl'
elif args.dataset in ["sbuilding"]:
    x_out=10
    args.data_dir='data/single_building.pkl'
elif args.dataset in ["smnist"]:
    x_out=10
    args.data_dir='data/single_mnist.pkl'
elif args.dataset in ["dbp"]:
    x_out=2
    args.data_dir='data/triple_building_600.pkl'
    
    
if args.model=="our":
    model=Smodel(h_channel=args.h_ch,input_featuresize=args.h_ch,\
                    localdepth=args.localdepth,num_interactions=args.num_interactions,finaldepth=args.finaldepth,share=args.share,batchnorm=args.batchnorm)
    mlpmodel=MLP(in_channels=args.h_ch*args.num_interactions, hidden_channels=args.h_ch,out_channels=x_out, num_layers=args.classifier_depth,dropout=args.dropout)

elif args.model=="HGT":
    model=model_new.HGT(hidden_channels=args.h_ch, out_channels=args.h_ch, num_heads=2, num_layers=args.num_interactions)
    mlpmodel=MLP(in_channels=args.h_ch, hidden_channels=args.h_ch,out_channels=x_out, num_layers=args.classifier_depth,dropout=args.dropout)
elif args.model=="HAN":
    model=model_new.HAN(hidden_channels=args.h_ch, out_channels=args.h_ch, num_heads=2, num_layers=args.num_interactions)
    mlpmodel=MLP(in_channels=args.h_ch, hidden_channels=args.h_ch,out_channels=x_out, num_layers=args.classifier_depth,dropout=args.dropout)

model.to(device), mlpmodel.to(device)
opt_list=list(model.parameters())+list(mlpmodel.parameters())

optimizer = torch.optim.Adam( opt_list, lr=args.lr)    
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, factor=0.1, patience=args.patience, min_lr=1e-8)   

def contrastive_loss(embeddings,labels,margin):
    
    positive_mask = labels.view(-1, 1) == labels.view(1, -1)
    negative_mask = ~positive_mask

    # Calculate the number of positive and negative pairs
    num_positive_pairs = positive_mask.sum() - labels.shape[0] 
    num_negative_pairs = negative_mask.sum()

    # If there are no negative pairs, return a placeholder loss
    if num_negative_pairs==0 or num_positive_pairs== 0:
        print("all pos or neg")
        return torch.tensor(0, dtype=torch.float)
    # Calculate the pairwise Euclidean distances between embeddings
    distances = torch.cdist(embeddings, embeddings)/np.sqrt(embeddings.shape[1])
    
    if num_positive_pairs>num_negative_pairs:
        # Sample an equal number of + pairs 
        positive_indices = torch.nonzero(positive_mask)
        random_positive_indices = torch.randperm(len(positive_indices))[:num_negative_pairs]
        selected_positive_indices = positive_indices[random_positive_indices]

        # Select corresponding negative pairs
        negative_mask.fill_diagonal_(False)
        negative_distances = distances[negative_mask].view(-1, 1)
        positive_distances = distances[selected_positive_indices[:,0],selected_positive_indices[:,1]].view(-1, 1)
    else: # case for most datasets
        # Sample an equal number of - pairs 
        negative_indices = torch.nonzero(negative_mask)
        random_negative_indices = torch.randperm(len(negative_indices))[:num_positive_pairs]
        selected_negative_indices = negative_indices[random_negative_indices]

        # Select corresponding positive pairs
        positive_mask.fill_diagonal_(False)
        positive_distances = distances[positive_mask].view(-1, 1)
        negative_distances = distances[selected_negative_indices[:,0],selected_negative_indices[:,1]].view(-1, 1)

    # Calculate the loss for positive and negative pairs
    loss = (positive_distances - negative_distances + margin).clamp(min=0).mean()
    return loss

def forward_HGT(data,model,mlpmodel):
    data = data.to(device)
    x,batch=data.pos, data['vertices'].batch
    data["vertices"]['x']=data.pos
    label=data.y.long().view(-1)

    optimizer.zero_grad()
    
    output=model(data.x_dict, data.edge_index_dict)  
    if args.dataset in ["dbp"]:
        graph_embeddings=global_add_pool(output,batch)
    else:
        graph_embeddings=global_add_pool(output,batch)
    graph_embeddings.clamp_(max=1e6)
    c_loss=contrastive_loss(graph_embeddings,label,margin=1)
    output=mlpmodel(graph_embeddings)
    # log_probs = F.log_softmax(output, dim=1)

    loss = criterion(output, label) 
    loss+=c_loss*args.loss_coef
    return loss,c_loss*args.loss_coef,output,label    
    
def forward(data,model,mlpmodel):
    data = data.to(device)
    edge_index1=data['vertices', 'inside', 'vertices']['edge_index']
    edge_index2=data['vertices', 'apart', 'vertices']['edge_index']
    combined_edge_index=torch.cat([data['vertices', 'inside', 'vertices']['edge_index'],data['vertices', 'apart', 'vertices']['edge_index']],1)
    num_edge_inside=edge_index1.shape[1]
    
    if args.spanning_tree == 'T':
        edge_weight=torch.rand(combined_edge_index.shape[1]) + 1
        undirected_spanning_edge = util.build_spanning_tree_edge(combined_edge_index, edge_weight,num_nodes=data.pos.shape[0])
        
        edge_set_1 = set(map(tuple, edge_index2.t().tolist()))
        edge_set_2 = set(map(tuple, undirected_spanning_edge.t().tolist()))

        common_edges = edge_set_1.intersection(edge_set_2)
        common_edges_tensor = torch.tensor(list(common_edges), dtype=torch.long).t().to(device)
        spanning_edge=torch.cat([edge_index1,common_edges_tensor],1)
        combined_edge_index=spanning_edge
    x,batch=data.pos, data['vertices'].batch
    label=data.y.long().view(-1)

    num_nodes=x.shape[0]
    edge_index_2rd, num_triplets_real, edx_jk, edx_ij = util.triplets(combined_edge_index, num_nodes)
    optimizer.zero_grad()
    input_feature=torch.zeros([x.shape[0],args.h_ch],device=device) 
    output=model(input_feature,x,[edge_index1,edge_index2], edge_index_2rd,edx_jk, edx_ij,batch,num_edge_inside,args.edge_rep)  
    output=torch.cat(output,dim=1)
    if args.dataset in ["dbp"]:
        graph_embeddings=global_add_pool(output,batch)
    else:
        graph_embeddings=global_add_pool(output,batch)
    graph_embeddings.clamp_(max=1e6)
    c_loss=contrastive_loss(graph_embeddings,label,margin=1)
    output=mlpmodel(graph_embeddings)

    loss = criterion(output, label) 
    loss+=c_loss*args.loss_coef
    return loss,c_loss*args.loss_coef,output,label
def train(train_Loader,model,mlpmodel ):
    epochloss=0
    epochcloss=0
    y_hat, y_true,y_hat_logit = [], [], []        
    optimizer.zero_grad()
    model.train()
    mlpmodel.train()
    for i,data in enumerate(train_Loader):
        if args.model=="our":
            loss,c_loss,output,label  =forward(data,model,mlpmodel)
        elif args.model in ["HGT","HAN"]:
            loss,c_loss,output,label  =forward_HGT(data,model,mlpmodel)

        loss.backward()
        optimizer.step()
        epochloss+=loss.detach().cpu()
        epochcloss+=c_loss.detach().cpu()
        
        _, pred = output.topk(1, dim=1, largest=True, sorted=True)
        pred,label,output=pred.cpu(),label.cpu(),output.cpu()
        y_hat += list(pred.detach().numpy().reshape(-1))
        y_true += list(label.detach().numpy().reshape(-1))
        y_hat_logit+=list(output.detach().numpy())
    return epochloss.item()/len(train_Loader),epochcloss.item()/len(train_Loader),y_hat, y_true,y_hat_logit

def test(loader,model,mlpmodel ):
    y_hat, y_true,y_hat_logit = [], [], []
    loss_total, pred_num = 0, 0
    model.eval()
    mlpmodel.eval()
    with torch.no_grad():
        for data in loader:
            if args.model=="our":
                loss,c_loss,output,label  =forward(data,model,mlpmodel)
            elif args.model in ["HGT","HAN"]:
                loss,c_loss,output,label  =forward_HGT(data,model,mlpmodel)
            
            _, pred = output.topk(1, dim=1, largest=True, sorted=True)
            pred,label,output=pred.cpu(),label.cpu(),output.cpu()
            y_hat += list(pred.detach().numpy().reshape(-1))
            y_true += list(label.detach().numpy().reshape(-1))
            y_hat_logit+=list(output.detach().numpy())
            
            pred_num += len(label.reshape(-1, 1))
            loss_total += loss.detach() * len(label.reshape(-1, 1))
    return loss_total/pred_num,y_hat, y_true, y_hat_logit
def main(args,train_Loader,val_Loader,test_Loader):
    best_val_trigger = -1
    old_lr=1e3
    suffix="{}{}-{}:{}:{}".format(datetime.now().strftime("%h"),
                                    datetime.now().strftime("%d"),
                                    datetime.now().strftime("%H"),
                                    datetime.now().strftime("%M"),
                                    datetime.now().strftime("%S"))        
    if args.log: writer = SummaryWriter(os.path.join(tensorboard_dir,suffix))

    for epoch in range(args.nepoch):
        train_loss,train_closs,y_hat, y_true,y_hat_logit=train(train_Loader,model,mlpmodel )

        train_acc=util.calculate(y_hat,y_true,y_hat_logit)
        try:util.record({"loss":train_loss,"closs":train_closs,"acc":train_acc},epoch,writer,"Train") 
        except: pass
        util.print_1(epoch,'Train',{"loss":train_loss,"closs":train_closs,"acc":train_acc})
        if epoch % args.test_per_round == 0:
            val_loss, yhat_val, ytrue_val, yhatlogit_val = test(val_Loader,model,mlpmodel )
            test_loss, yhat_test, ytrue_test, yhatlogit_test = test(test_Loader,model,mlpmodel )
            val_acc=util.calculate(yhat_val,ytrue_val,yhatlogit_val)
            try:util.record({"loss":val_loss,"acc":val_acc},epoch,writer,"Val")
            except: pass
            util.print_1(epoch,'Val',{"loss":val_loss,"acc":val_acc},color=blue) 
            test_acc=util.calculate(yhat_test,ytrue_test,yhatlogit_test)
            try:util.record({"loss":test_loss,"acc":test_acc},epoch,writer,"Test")            
            except: pass
            util.print_1(epoch,'Test',{"loss":test_loss,"acc":test_acc},color=blue)
            val_trigger=val_acc
            if val_trigger > best_val_trigger:
                best_val_trigger = val_trigger
                best_model = copy.deepcopy(model)
                best_mlpmodel=copy.deepcopy(mlpmodel)
                best_info=[epoch,val_trigger]
        """ 
        update lr when epoch≥30
        """
        if epoch >= 30:
            lr = scheduler.optimizer.param_groups[0]['lr']
            if old_lr!=lr:
                print(red('lr'), epoch, (lr), sep=', ')
                old_lr=lr
            scheduler.step(val_trigger)        
    """
    use best model to get best model result 
    """
    val_loss, yhat_val, ytrue_val, yhat_logit_val  = test(val_Loader,best_model,best_mlpmodel)
    test_loss, yhat_test, ytrue_test, yhat_logit_test= test(test_Loader,best_model,best_mlpmodel)

    val_acc=util.calculate(yhat_val,ytrue_val,yhat_logit_val)
    util.print_1(best_info[0],'BestVal',{"loss":val_loss,"acc":val_acc},color=blue)
    test_acc=util.calculate(yhat_test,ytrue_test,yhat_logit_test)
    util.print_1(best_info[0],'BestTest',{"loss":test_loss,"acc":test_acc},color=blue)
    if args.model=="our":print(best_model.att)
                                                            
    """
    save training info and best result 
    """
    result_file=os.path.join(info_dir, suffix)
    with open(result_file, 'w') as f:
        print("Random Seed: ", Seed,file=f)
        print(f"acc  val : {val_acc:.3f}, Test : {test_acc:.3f}", file=f)
        print(f"Best info: {best_info}", file=f)
        for i in [[a,getattr(args, a)] for a in args.__dict__]:
            print(i,sep='\n',file=f)
    to_save_dict={'model':best_model.state_dict(),'mlpmodel':best_mlpmodel.state_dict(),'args':args,'labels':ytrue_test,'yhat':yhat_test,'yhat_logit':yhat_logit_test}
    torch.save(to_save_dict, os.path.join(model_dir,suffix+'.pth') )
    print("done")

if __name__ == '__main__':
    """
    build dir 
    """
    if not os.path.exists(args.save_dir):
        os.makedirs(args.save_dir,exist_ok=True)
    tensorboard_dir=os.path.join(args.save_dir,'log')
    if not os.path.exists(tensorboard_dir):
        os.makedirs(tensorboard_dir,exist_ok=True)
    model_dir=os.path.join(args.save_dir,'model')
    if not os.path.exists(model_dir):
        os.makedirs(model_dir,exist_ok=True)    
    info_dir=os.path.join(args.save_dir,'info')
    if not os.path.exists(info_dir):
        os.makedirs(info_dir,exist_ok=True)      

    Seed = 0
    test_ratio=0.2
    print("data splitting Random Seed: ", Seed)
    if args.dataset in ['mnist',"mnist_sparse"]:
        train_ds,val_ds,test_ds=dataset.get_mnist_dataset(args.data_dir,Seed,test_ratio=test_ratio)
    elif args.dataset in ['building']:
        train_ds,val_ds,test_ds=dataset.get_building_dataset(args.data_dir,Seed,test_ratio=test_ratio)
    elif args.dataset in ['mbuilding']:
        train_ds,val_ds,test_ds=dataset.get_mbuilding_dataset(args.data_dir,Seed,test_ratio=test_ratio)
    elif args.dataset in ['sbuilding']:
        train_ds,val_ds,test_ds=dataset.get_sbuilding_dataset(args.data_dir,Seed,test_ratio=test_ratio)
    elif args.dataset in ['smnist']:
        train_ds,val_ds,test_ds=dataset.get_smnist_dataset(args.data_dir,Seed,test_ratio=test_ratio)    
    elif args.dataset in ['dbp']:
        train_ds,val_ds,test_ds=dataset.get_dbp_dataset(args.data_dir,Seed,test_ratio=test_ratio)
    if args.extent_norm=="T":
        train_ds= dataset.affine_transform_to_range(train_ds,target_range=(-1, 1))
        val_ds= dataset.affine_transform_to_range(val_ds,target_range=(-1, 1))
        test_ds= dataset.affine_transform_to_range(test_ds,target_range=(-1, 1))
                
    train_loader = torch_geometric.loader.DataLoader(train_ds,batch_size=args.train_batch, shuffle=False,pin_memory=True,drop_last=True) 
    val_loader = torch_geometric.loader.DataLoader(val_ds, batch_size=args.test_batch, shuffle=False, pin_memory=True)
    test_loader = torch_geometric.loader.DataLoader(test_ds,batch_size=args.test_batch, shuffle=False,pin_memory=True)
    """
    set model seed 
    """
    Seed = args.man_seed if args.manualSeed else random.randint(1, 10000)
    Seed=3407
    print("Random Seed: ", Seed)
    print(args)
    random.seed(Seed)
    torch.manual_seed(Seed)  
    np.random.seed(Seed)    
    main(args,train_loader,val_loader,test_loader)