Spaces:
Sleeping
Sleeping
Hong
commited on
Commit
·
7e9a171
1
Parent(s):
b1026c7
Upload BART_utils.py
Browse files- BART_utils.py +101 -0
BART_utils.py
ADDED
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
from load_data import *
|
3 |
+
import matplotlib.pyplot as plt
|
4 |
+
import streamlit as st
|
5 |
+
import torch
|
6 |
+
|
7 |
+
from transformers import AutoTokenizer
|
8 |
+
from transformers import AutoModelForSequenceClassification
|
9 |
+
|
10 |
+
|
11 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
12 |
+
|
13 |
+
tokenizer = AutoTokenizer.from_pretrained("facebook/bart-large-mnli")
|
14 |
+
nli_model = (
|
15 |
+
AutoModelForSequenceClassification.from_pretrained(
|
16 |
+
"facebook/bart-large-mnli"
|
17 |
+
).cuda()
|
18 |
+
if torch.cuda.is_available()
|
19 |
+
else AutoModelForSequenceClassification.from_pretrained("facebook/bart-large-mnli")
|
20 |
+
)
|
21 |
+
|
22 |
+
|
23 |
+
def get_prob(sequence, label):
|
24 |
+
premise = sequence
|
25 |
+
hypothesis = f"This example is {label}."
|
26 |
+
|
27 |
+
# run through model pre-trained on MNLI
|
28 |
+
x = tokenizer.encode(
|
29 |
+
premise, hypothesis, return_tensors="pt", truncation_strategy="only_first"
|
30 |
+
)
|
31 |
+
logits = nli_model(x.to(device))[0]
|
32 |
+
|
33 |
+
# we throw away "neutral" (dim 1) and take the probability of
|
34 |
+
# "entailment" (2) as the probability of the label being true
|
35 |
+
entail_contradiction_logits = logits[:, [0, 2]]
|
36 |
+
probs = entail_contradiction_logits.softmax(dim=1)
|
37 |
+
prob_label_is_true = probs[:, 1]
|
38 |
+
return prob_label_is_true[0].item()
|
39 |
+
|
40 |
+
|
41 |
+
def judge_mbti(sequence, labels):
|
42 |
+
out = []
|
43 |
+
for l in labels:
|
44 |
+
temp = get_prob(sequence, l)
|
45 |
+
out.append((l, temp))
|
46 |
+
out = sorted(out, key=lambda x: x[1], reverse=True)
|
47 |
+
return out
|
48 |
+
|
49 |
+
|
50 |
+
def compute_score(text, type):
|
51 |
+
x, y = type.split("_")
|
52 |
+
x_score = np.sum([i[1] for i in judge_mbti(text, keywords_en[type][x])])
|
53 |
+
y_score = np.sum([i[1] for i in judge_mbti(text, keywords_en[type][y])])
|
54 |
+
|
55 |
+
if x_score > y_score:
|
56 |
+
choice = x
|
57 |
+
score = x_score
|
58 |
+
else:
|
59 |
+
choice = y
|
60 |
+
score = y_score
|
61 |
+
|
62 |
+
x_score_scaled = (x_score / (x_score + y_score)) * 100
|
63 |
+
y_score_scaled = (y_score / (x_score + y_score)) * 100
|
64 |
+
|
65 |
+
stat = {x: x_score_scaled, y: y_score_scaled}
|
66 |
+
|
67 |
+
return choice, stat
|
68 |
+
|
69 |
+
|
70 |
+
def mbti_translator(text):
|
71 |
+
E_I = compute_score(text, "E_I")
|
72 |
+
N_S = compute_score(text, "N_S")
|
73 |
+
T_F = compute_score(text, "T_F")
|
74 |
+
P_J = compute_score(text, "P_J")
|
75 |
+
|
76 |
+
return (E_I[0] + N_S[0] + T_F[0] + P_J[0]), (E_I[1], N_S[1], T_F[1], P_J[1])
|
77 |
+
|
78 |
+
|
79 |
+
def plot_mbti(result):
|
80 |
+
fig, ax = plt.subplots(figsize=(10, 5))
|
81 |
+
|
82 |
+
start = 0
|
83 |
+
x, y = result.values()
|
84 |
+
x_type, y_type = result.keys()
|
85 |
+
|
86 |
+
ax.broken_barh([(start, x), (x, x + y)], [10, 9],
|
87 |
+
facecolors=("#FFC5BF", "#D4F0F0"))
|
88 |
+
ax.set_ylim(5, 15)
|
89 |
+
ax.set_xlim(0, 100)
|
90 |
+
ax.spines["left"].set_visible(False)
|
91 |
+
ax.spines["bottom"].set_visible(False)
|
92 |
+
ax.spines["top"].set_visible(False)
|
93 |
+
ax.spines["right"].set_visible(False)
|
94 |
+
|
95 |
+
ax.set_yticks([15, 25])
|
96 |
+
ax.set_xticks([0, 25, 50, 75, 100])
|
97 |
+
|
98 |
+
ax.text(x - 6, 14.5, x_type + " :" + str(int(x)) + "%", fontsize=15)
|
99 |
+
ax.text((x + y) - 6, 14.5, y_type + " :" + str(int(y)) + "%", fontsize=15)
|
100 |
+
|
101 |
+
st.pyplot(fig)
|