forked from Yichuan0712/11785-TCR
-
Notifications
You must be signed in to change notification settings - Fork 0
/
extract.py
183 lines (143 loc) · 7.74 KB
/
extract.py
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
import pickle
from util import printl
import torch
import numpy as np
from tqdm import tqdm
import os
import torch.nn.functional as F
import pandas as pd
from scipy.stats import skew, kurtosis
def extract_features(encoder, projection_head, train1_loader, train2_loader, test_loader, tokenizer, log_path):
device = torch.device("cuda")
encoder.eval()
projection_head.eval()
progress_bar = tqdm(enumerate(train1_loader), total=len(train1_loader), desc=f"Cluster Center Calculation")
log_dir = os.path.dirname(log_path)
log_file_average = os.path.join(log_dir, "epitope_average.pkl")
epitope_sums = {}
epitope_counts = {}
with torch.no_grad():
for batch, data in progress_bar:
epitope_list = data['epitope']
anchor_list = data['TCR']
anchor_seq_batch = [(epitope_list[i], str(anchor_list[i])) for i in range(len(epitope_list))]
_, _, anchor_tokens = tokenizer(anchor_seq_batch)
anchor_embs = projection_head(encoder(anchor_tokens.to(device)).mean(dim=1))
for i, epitope in enumerate(epitope_list):
if epitope not in epitope_sums:
epitope_sums[epitope] = anchor_embs[i]
epitope_counts[epitope] = 1
else:
epitope_sums[epitope] += anchor_embs[i]
epitope_counts[epitope] += 1
epitope_data = {
epitope: {
"average_embedding": (epitope_sums[epitope] / epitope_counts[epitope]),
"count": epitope_counts[epitope]
}
for epitope in epitope_sums
}
with open(log_file_average, "wb") as f:
pickle.dump(epitope_data, f)
printl(f"Cluster center calculation completed and saved to {log_file_average}.", log_path=log_path)
"""train2"""
progress_bar2 = tqdm(enumerate(train2_loader), total=len(train2_loader), desc="Finding Nearest Cluster Centers")
true_classes = []
feature_list = []
with torch.no_grad():
for batch, data in progress_bar2:
epitope_list = data['epitope']
anchor_list = data['TCR']
label_list = data['label']
anchor_seq_batch = [(epitope_list[i], str(anchor_list[i])) for i in range(len(epitope_list))]
_, _, anchor_tokens = tokenizer(anchor_seq_batch)
anchor_embs = projection_head(encoder(anchor_tokens.to(device)).mean(dim=1))
for i, epitope in enumerate(epitope_list):
true_classes.append(epitope)
cosine_similarities = []
for cluster_epitope, cluster_data in epitope_data.items():
cluster_emb = cluster_data["average_embedding"].to(device)
cos_sim = F.cosine_similarity(anchor_embs[i].unsqueeze(0), cluster_emb.unsqueeze(0)).item()
cosine_similarities.append((cluster_epitope, cos_sim))
cosine_similarities.sort(key=lambda x: x[1])
similarity_values = [d[1] for d in cosine_similarities]
min_similarity = min(similarity_values)
max_similarity = max(similarity_values)
avg_similarity = sum(similarity_values) / len(similarity_values)
median_similarity = np.median(similarity_values)
std_similarity = np.std(similarity_values)
skewness_similarity = skew(similarity_values)
kurtosis_similarity = kurtosis(similarity_values, fisher=True)
target_cluster_emb = epitope_data[epitope]["average_embedding"].to(device)
similarity_to_own_cluster = F.cosine_similarity(anchor_embs[i].unsqueeze(0), target_cluster_emb.unsqueeze(0)).item()
rank_position = [d[0] for d in cosine_similarities].index(epitope) + 1 # 索引从0开始，故加1
features = {
'x': anchor_list[i],
'y': epitope,
'similarity_to_own_cluster': similarity_to_own_cluster,
'max_similarity': max_similarity,
'min_similarity': min_similarity,
'avg_similarity': avg_similarity,
'median_similarity': median_similarity,
'std_similarity': std_similarity,
'skewness_similarity': skewness_similarity,
'kurtosis_similarity': kurtosis_similarity,
'rank_position': rank_position,
'label': int(label_list[i]),
}
feature_list.append(features)
feature_df = pd.DataFrame(feature_list)
csv_path = os.path.join(log_dir, 'feature_data_train.csv')
feature_df.to_csv(csv_path, index=False)
printl(f"Training features are saved to {csv_path}.", log_path=log_path)
"""test"""
progress_bar3 = tqdm(enumerate(test_loader), total=len(test_loader), desc="Finding Nearest Cluster Centers")
true_classes = []
feature_list = []
with torch.no_grad():
for batch, data in progress_bar3:
epitope_list = data['epitope']
anchor_list = data['TCR']
label_list = data['label']
anchor_seq_batch = [(epitope_list[i], str(anchor_list[i])) for i in range(len(epitope_list))]
_, _, anchor_tokens = tokenizer(anchor_seq_batch)
anchor_embs = projection_head(encoder(anchor_tokens.to(device)).mean(dim=1))
for i, epitope in enumerate(epitope_list):
true_classes.append(epitope)
cosine_similarities = []
for cluster_epitope, cluster_data in epitope_data.items():
cluster_emb = cluster_data["average_embedding"].to(device)
cos_sim = F.cosine_similarity(anchor_embs[i].unsqueeze(0), cluster_emb.unsqueeze(0)).item()
cosine_similarities.append((cluster_epitope, cos_sim))
cosine_similarities.sort(key=lambda x: x[1])
similarity_values = [d[1] for d in cosine_similarities]
min_similarity = min(similarity_values)
max_similarity = max(similarity_values)
avg_similarity = sum(similarity_values) / len(similarity_values)
median_similarity = np.median(similarity_values)
std_similarity = np.std(similarity_values)
skewness_similarity = skew(similarity_values)
kurtosis_similarity = kurtosis(similarity_values, fisher=True)
target_cluster_emb = epitope_data[epitope]["average_embedding"].to(device)
similarity_to_own_cluster = F.cosine_similarity(anchor_embs[i].unsqueeze(0),
target_cluster_emb.unsqueeze(0)).item()
rank_position = [d[0] for d in cosine_similarities].index(epitope) + 1 # 索引从0开始，故加1
features = {
'x': anchor_list[i],
'y': epitope,
'similarity_to_own_cluster': similarity_to_own_cluster,
'max_similarity': max_similarity,
'min_similarity': min_similarity,
'avg_similarity': avg_similarity,
'median_similarity': median_similarity,
'std_similarity': std_similarity,
'skewness_similarity': skewness_similarity,
'kurtosis_similarity': kurtosis_similarity,
'rank_position': rank_position,
'label': int(label_list[i]),
}
feature_list.append(features)
feature_df = pd.DataFrame(feature_list)
csv_path = os.path.join(log_dir, 'feature_data_test.csv')
feature_df.to_csv(csv_path, index=False)
printl(f"Test features are saved to {csv_path}.", log_path=log_path)