-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathevaluation.py
204 lines (180 loc) · 7.34 KB
/
evaluation.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
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
import numpy as np
import pandas as pd
import torch
import utils
from tqdm import tqdm
import transformers
from mlm_simple.models.BERTquery import MLMBERT
from sklearn.model_selection import train_test_split
# [(id, seq), ...]
def single_mask_test(models, queries, config, model_names=None, outpath=None):
if model_names is None:
model_names=np.arange(len(models))
if len(model_names) != len(models):
print('len(model_names) != models. setting model names to index in list')
model_names=np.arange(len(models))
np.random.seed(0)
mask_indices=[np.random.randint(0,len(seq)) for seq in queries['Sequence']]
labels=[(id, seq) for i, (id, seq) in queries.iterrows()]
batch = [
(row["Header"], utils.mask(row["Sequence"], index))
for (index, row), index in zip(queries.iterrows(), mask_indices)
]
tokenizer = transformers.AutoTokenizer.from_pretrained(
'mlm_simple/mlm-baby-bert/tokenizer/protein_tokenizer',
use_fast=True,
unk_token="<unk>",
mask_token="<mask>",
pad_token="<pad>"
)
batch_ids=torch.Tensor([tokenizer.encode(seq)+(config['max_len']-len(tokenizer.encode(seq)))*[config['pad_token_id']] for _,seq in batch if len(seq)<1000]).int()
results_list=[]
for model, name in zip(models,model_names):
# Inference
with torch.no_grad():
results = model(batch_ids)['logits'].argmax(dim=-1)
# Evaluation
for pred,(header, seq),i in tqdm(zip(results, labels, mask_indices)):
results_list.append({'Name': name,'Header': header, 'Sequence': seq,
'Mask': i, 'Prediction': tokenizer.decode(pred[i+1]),
'Label': seq[i], 'Correct': int(seq[i]==tokenizer.decode(pred[i+1]))})
results_df=pd.DataFrame(results_list)
if outpath is not None:
results_df.to_csv(outpath)
return results_df
def single_mask_test_MSA(models, queries, config, model_names=None, outpath=None):
if model_names is None:
model_names=np.arange(len(models))
if len(model_names) != len(models):
print('len(model_names) != len(models). setting model names to index in list')
model_names=np.arange(len(models))
tokenizer = transformers.AutoTokenizer.from_pretrained(
'mlm_simple/mlm-baby-bert/tokenizer/protein_tokenizer',
use_fast=True,
unk_token="<unk>",
mask_token="<mask>",
pad_token="<pad>"
)
np.random.seed(0)
#[print(head.split('|')[1]) for i,(head,seq,query) in queries.iterrows()]
mask_indices=[np.random.randint(0,len(MSA['MSA'][0])) for i, (id, MSA) in queries.iterrows()]
labels=[(id, ['MSA'][0]) for i, (id, MSA) in queries.iterrows()]
print(labels)
batch = [(tokenizer.encode(list(utils.mask_MSA(['MSA'][0], index)))+(config['max_len']-len(tokenizer.encode(seq)))*[config['pad_token_id']], ['MSA'][1]) for i, (id, MSA) in queries.iterrows()]
results_list=[]
for model, name in zip(models,model_names):
# Inference
with torch.no_grad():
results = model(batch)['logits'].argmax(dim=-1)
# Evaluation
for pred,(header, seq),i in tqdm(zip(results, labels, mask_indices)):
results_list.append({'Name': name,'Header': header, 'Sequence': seq,
'Mask': i, 'Prediction': tokenizer.decode(pred[i+1]),
'Label': seq[i], 'Correct': int(seq[i]==tokenizer.decode(pred[i+1]))})
results_df=pd.DataFrame(results_list)
if outpath is not None:
results_df.to_csv(outpath)
return results_df
def load_model(filepath, config):
model = MLMBERT(config).to(config['device'])
state_dict = torch.load(filepath)
model.load_state_dict(state_dict, strict=False)
return model
if __name__ == "__main__":
np.random.seed(0)
queries=pd.read_csv('input/all_queries.csv')
_,queries=train_test_split(queries, test_size=0.3, random_state=42, shuffle=True)
_,queries=train_test_split(queries, test_size=0.5, random_state=42, shuffle=True)
print('data imported')
tokenizer = transformers.AutoTokenizer.from_pretrained(
'mlm_simple/mlm-baby-bert/tokenizer/protein_tokenizer',
use_fast=True,
unk_token="<unk>",
mask_token="<mask>",
pad_token="<pad>"
)
configs = [
{
'batch_size': 32,
'dim': 256,
'n_heads': 8,
'attn_dropout': 0.1,
'mlp_dropout': 0.1,
'depth': 6,
'vocab_size': len(tokenizer.get_vocab()),
'max_len': 1000,
'pad_token_id': tokenizer.pad_token_id,
'mask_token_id': tokenizer.mask_token_id,
'device': 'cpu'
},
{
'batch_size': 32,
'dim': 512,
'n_heads': 8,
'attn_dropout': 0.1,
'mlp_dropout': 0.1,
'depth': 4,
'vocab_size': len(tokenizer.get_vocab()),
'max_len': 1000,
'pad_token_id': tokenizer.pad_token_id,
'mask_token_id': tokenizer.mask_token_id,
'device': 'cpu'
},
{
'batch_size': 32,
'dim': 256,
'n_heads': 16,
'attn_dropout': 0.1,
'mlp_dropout': 0.1,
'depth': 4,
'vocab_size': len(tokenizer.get_vocab()),
'max_len': 1000,
'pad_token_id': tokenizer.pad_token_id,
'mask_token_id': tokenizer.mask_token_id,
'device': 'cpu'
},
{
'batch_size': 32,
'dim': 256,
'n_heads': 8,
'attn_dropout': 0.1,
'mlp_dropout': 0.1,
'depth': 4,
'vocab_size': len(tokenizer.get_vocab()),
'max_len': 1000,
'pad_token_id': tokenizer.pad_token_id,
'mask_token_id': tokenizer.mask_token_id,
'device': 'cpu'
}
]
filepaths = [
"mlm_simple/mlm-baby-bert/BERT_depth4_embed256_steps2000.pt",
"mlm_simple/mlm-baby-bert/BERT_depth4_embed512_steps2000.pt",
"mlm_simple/mlm-baby-bert/BERT_depth4_embed256_head16_steps1000.pt",
'mlm_simple/mlm-baby-bert/BERTMSA_depth4_embed256_head8_steps2000.pt'
]
models=[load_model(filepath, config) for config,filepath in zip(configs, filepaths)]
print('models imported')
names=[filepath.split('/')[-1].split('.')[0] for filepath in filepaths]
queries=queries[queries['Sequence'].apply(len)<configs[0]['max_len']][:100]
res=single_mask_test(models, queries, configs[0], names, outpath='model_evaluation.csv')
np.random.seed(0)
tokenizer = transformers.AutoTokenizer.from_pretrained(
'mlm_simple/mlm-baby-bert/tokenizer/protein_tokenizer',
use_fast=True,
unk_token="<unk>",
mask_token="<mask>",
pad_token="<pad>")
config={
'batch_size': 32,
'dim': 256,
'n_heads': 8,
'attn_dropout': 0.1,
'mlp_dropout': 0.1,
'depth': 4,
'vocab_size': len(tokenizer.get_vocab()),
'max_len': 1000,
'pad_token_id': tokenizer.pad_token_id,
'mask_token_id': tokenizer.mask_token_id,
'device': 'cpu'}
res=single_mask_test(models, queries, config, names, outpath='model_evaluation.csv')