[c3444c]: / test / cadec / cadec_eval.py

Download this file

182 lines (148 with data), 6.1 kB

  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
from gensim import models
import os
import sys
import torch
import numpy as np
from transformers import AutoTokenizer, AutoModel, AutoConfig
batch_size = 64
device = "cuda:1"
def main():
filename = sys.argv[1]
print(filename)
bert_like = False
if filename[-3:] in ["vec", "txt"]:
W = load_vectors(filename, dev=False)
elif filename[-3:] == "bin":
W = load_vectors_bin(filename)
else:
bert_like = True
try:
config = AutoConfig.from_pretrained(filename)
model = AutoModel.from_pretrained(
filename, config=config).to(device)
except BaseException:
model = torch.load(os.path.join(
filename, 'pytorch_model.bin')).to(device)
try:
model.output_hidden_states = False
except BaseException:
pass
try:
tokenizer = AutoTokenizer.from_pretrained(filename)
except BaseException:
tokenizer = AutoTokenizer.from_pretrained(
os.path.join(filename, "../"))
top_k = 3
if bert_like:
eval(model, tokenizer, './cadec/data/cadec', top_k=top_k, summary_method="CLS")
eval(model, tokenizer, './cadec/data/cadec', top_k=top_k, summary_method="MEAN")
eval(model, tokenizer, './cadec/data/psytar_disjoint_folds', top_k=top_k, summary_method="CLS")
eval(model, tokenizer, './cadec/data/psytar_disjoint_folds', top_k=top_k, summary_method="MEAN")
else:
eval(W, None, './cadec/data/cadec', top_k=top_k)
eval(W, None, './cadec/data/psytar_disjoint_folds', top_k=top_k)
def get_bert_embed(phrase_list, m, tok, normalize=True, summary_method="CLS"):
input_ids = []
for phrase in phrase_list:
input_ids.append(tok.encode_plus(
phrase, max_length=32, add_special_tokens=True,
truncation=True, pad_to_max_length=True)['input_ids'])
m.eval()
count = len(input_ids)
now_count = 0
with torch.no_grad():
while now_count < count:
input_gpu_0 = torch.LongTensor(input_ids[now_count:min(
now_count + batch_size, count)]).to(device)
if summary_method == "CLS":
embed = m(input_gpu_0)[1]
if summary_method == "MEAN":
embed = torch.mean(m(input_gpu_0)[0], dim=1)
if normalize:
embed_norm = torch.norm(
embed, p=2, dim=1, keepdim=True).clamp(min=1e-12)
embed = embed / embed_norm
embed_np = embed.cpu().detach().numpy()
if now_count == 0:
output = embed_np
else:
output = np.concatenate((output, embed_np), axis=0)
now_count = min(now_count + batch_size, count)
return output
def eval_one(m, tok, folder, top_k, summary_method=None):
with open(os.path.join(folder, "standard.txt"), "r", encoding="utf-8") as f:
lines = f.readlines()
label2id = {line.strip().split(
"\t")[0]: index for index, line in enumerate(lines)}
standard_lines = [line.strip().split("\t") for line in lines]
#standard_feat = np.array([get_bert_embed(text, m, tok) for (label, text) in standard_lines])
if tok is not None:
standard_feat = get_bert_embed(
[text for (label, text) in standard_lines], m, tok, normalize=True, summary_method=summary_method)
else:
standard_feat = embed(
[text for (label, text) in standard_lines], m.vector_size, m)
with open(os.path.join(folder, "test.txt"), "r", encoding="utf-8") as f:
lines = f.readlines()
test_lines = [line.strip().split("\t") for line in lines]
#test_feat = np.array([get_bert_embed(text, m, tok) for (label, text) in test_lines])
if tok is not None:
test_feat = get_bert_embed(
[text for (label, text) in test_lines], m, tok, normalize=True, summary_method=summary_method)
else:
test_feat = embed(
[text for (label, text) in test_lines], m.vector_size, m)
sim_mat = np.dot(test_feat, standard_feat.T)
correct_1 = 0
correct_k = 0
pred_top_k = torch.topk(torch.FloatTensor(sim_mat), k=top_k)[
1].cpu().numpy()
for i in range(len(test_lines)):
true_id = label2id[test_lines[i][0]]
if pred_top_k[i][0] == true_id:
correct_1 += 1
if true_id in list(pred_top_k[i]):
correct_k += 1
acc_1 = correct_1 / len(test_lines)
acc_k = correct_k / len(test_lines)
return acc_1, acc_k
def eval(m, tok, task_name, top_k=3, summary_method=None):
acc_1_list = []
acc_k_list = []
for p in os.listdir(task_name):
acc_1, acc_k = eval_one(m, tok, os.path.join(task_name, p), top_k, summary_method=summary_method)
acc_1_list.append(acc_1)
acc_k_list.append(acc_k)
print(task_name, summary_method)
print(f"top_k={top_k}")
print(acc_1_list)
print(acc_k_list)
print(sum(acc_1_list) / 5, sum(acc_k_list) / 5)
return None
def load_vectors(filename):
W = {}
with open(filename, 'r') as f:
for i, line in enumerate(f.readlines()):
if i == 0:
continue
toks = line.strip().split()
w = toks[0]
vec = np.array(map(float, toks[1:]))
W[w] = vec
return W
def load_vectors_bin(filename):
w = models.KeyedVectors.load_word2vec_format(filename, binary=True)
return w
def cosine(u, v):
return np.dot(u, v)
def norm(v):
return np.dot(v, v)**0.5
def embed_one(phrase, dim, W):
words = phrase.split()
vectors = [W[w] for w in words if (w in W)]
v = sum(vectors, np.zeros(dim))
return v / (norm(v) + 1e-9)
def embed(phrase_list, dim, W):
return np.array([embed_one(phrase, dim, W) for phrase in phrase_list])
if __name__ == '__main__':
main()