-
Notifications
You must be signed in to change notification settings - Fork 0
/
translate_ldc.py
215 lines (170 loc) · 8.79 KB
/
translate_ldc.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
204
205
206
207
208
209
210
211
212
213
214
215
# For NIST dataset only
# Modified from generate.py
# Translate diverse sentences from the input stream (appended with codes at beginning) and then evaluate.
#
# The model will be faster is sentences are sorted by length.
# Input sentences must have the same tokenization and BPE codes than the ones used in the model.
import os
import io
import sys
import argparse
import torch
import numpy as np
from src.utils import AttrDict
from src.utils import initialize_exp, restore_segmentation
from src.data.dictionary import Dictionary
from src.model.transformer import TransformerModel
from src.evaluation.evaluator import convert_to_text, convert_to_diverse_text, eval_mbleu, eval_diverse_bleu
from src.data.loader import set_dico_parameters
from src.data.dataset import Dataset, ParallelDataset
from src.utils import to_cuda
from logging import getLogger
logger = getLogger()
def get_parser():
"""
Generate a parameters parser.
"""
# parse parameters
parser = argparse.ArgumentParser(description="Translate sentences")
# main parameters
parser.add_argument("--dump_path", type=str, default="/dumped/", help="Experiment dump path")
parser.add_argument("--exp_name", type=str, default="", help="Experiment name")
parser.add_argument("--exp_id", type=str, default="", help="Experiment ID")
parser.add_argument("--batch_size", type=int, default=32, help="Number of sentences per batch")
parser.add_argument("--tokens_per_batch", type=int, default=6250, help="")
parser.add_argument("--max_batch_size", type=int, default=0,
help="Maximum number of sentences per batch (used in combination with tokens_per_batch, 0 to disable)")
# model / output / reference / src paths
parser.add_argument("--model_path", type=str, default="", help="Model path")
parser.add_argument("--dataset", type=str, default="ldc-zhen-codes/", help="Dataset path")
# source language / target language
parser.add_argument("--src_lang", type=str, default="", help="Source language")
parser.add_argument("--tgt_lang", type=str, default="", help="Target language")
parser.add_argument("--k", type=int, default=3, help="Generate top k")
parser.add_argument("--beam_size", type=int, default=5, help="Beam size")
return parser
def main(params):
# initialize the experiment
logger = initialize_exp(params)
assert os.path.isdir(params.dump_path)
# generate parser / parse parameters
parser = get_parser()
params = parser.parse_args()
reloaded = torch.load(params.model_path)
model_params = AttrDict(reloaded['params'])
logger.info("Supported languages: %s" % ", ".join(model_params.lang2id.keys()))
params.src_id = model_params.lang2id[params.src_lang]
params.tgt_id = model_params.lang2id[params.tgt_lang]
# Load a binarized dataset.
def load_binarized(data_path):
assert data_path.endswith('.pth')
assert os.path.isfile(data_path), data_path
logger.info("Loading data from %s ..." % data_path)
data = torch.load(data_path)
# process_binarized
dico = data['dico']
assert ((data['sentences'].dtype == np.uint16) and (len(dico) < 1 << 16) or
(data['sentences'].dtype == np.int32) and (1 << 16 <= len(dico) < 1 << 31))
logger.info("%i words (%i unique) in %i sentences. %i unknown words (%i unique) covering %.2f%% of the data." % (
len(data['sentences']) - len(data['positions']),
len(dico), len(data['positions']),
sum(data['unk_words'].values()), len(data['unk_words']),
100. * sum(data['unk_words'].values()) / (len(data['sentences']) - len(data['positions']))
))
if (data['sentences'].dtype == np.int32) and (len(dico) < 1 << 16):
logger.info("Less than 65536 words. Moving data from int32 to uint16 ...")
data['sentences'] = data['sentences'].astype(np.uint16)
# update dictionary parameters
set_dico_parameters(params, data, data['dico']) # will params be updated?
return data
root_path = "/data/"
data_root = root_path+"ldc"
encoder = None
decoder = None
for nist in ["nist02", "nist03", "nist04", "nist05", "nist06", "nist08"]:
src_path = root_path + params.dataset + nist + ".src.bpe.pth"
ref_path = root_path + params.dataset + nist + ".ref0.bpe.pth" # will not be directly used
src_data = load_binarized(src_path)
tgt_data = load_binarized(ref_path)
#dataset = Dataset(data['sentences'], data['positions'], params)
dataset = ParallelDataset(
src_data['sentences'], src_data['positions'],
tgt_data['sentences'], tgt_data['positions'],
params
)
if encoder is None:
# build dictionary / build encoder / build decoder / reload weights
encoder = TransformerModel(model_params, src_data['dico'], is_encoder=True, with_output=True).cuda().eval()
decoder = TransformerModel(model_params, tgt_data['dico'], is_encoder=False, with_output=True).cuda().eval()
encoder.load_state_dict({k.replace('module.',''):v for k,v in reloaded['encoder'].items()})
decoder.load_state_dict({k.replace('module.',''):v for k,v in reloaded['decoder'].items()})
iterator = dataset.get_iterator(
shuffle=False,
group_by_size=False,
n_sentences=-1,)
K = params.k
hypothesis_ks = []
for i in range(K):
hypothesis_ks.append([])
lang2_txt = []
mean_conf = torch.zeros(K)
for batch in iterator:
if len(lang2_txt)>200:
break
(x1, len1), (x2, len2) = batch
# if train
# x1 = torch.cat((x1[0,:].unsqueeze(0),x1[2:,:])) #remove the codes in src
# len1-=1
langs1 = x1.clone().fill_(params.src_id)
langs2 = x2.clone().fill_(params.tgt_id)
# cuda
x1, len1, langs1, x2, len2, langs2 = to_cuda(x1, len1, langs1, x2, len2, langs2)
# encode source sentence
enc1 = encoder('fwd', x=x1, lengths=len1, langs=langs1, causal=False)
enc1 = enc1.transpose(0, 1)
max_len = int(1.5 * len1.max().item() + 10)
generated_ks, lengths_ks, conf_ks = decoder.generate_diverse_beam(
K, enc1, len1, params.tgt_id, beam_size=params.beam_size,
length_penalty=1,
early_stopping=False,
max_len=max_len)
assert len(generated_ks)==K
assert len(lengths_ks)==K
mean_conf = (mean_conf + conf_ks.cpu()) / 2.0
for i in range(K):
hypothesis_ks[i].extend(convert_to_diverse_text(generated_ks[i], lengths_ks[i], tgt_data['dico'], params))
#hypothesis_ks[i].extend(convert_to_text(generated_ks[i], lengths_ks[i], tgt_data['dico'], params))
'''
hyp_bin = "{0}_hyp_{1}-{2}.K0.pth".format(nist, params.src_lang, params.tgt_lang)
hyp_bin_path = os.path.join(params.dump_path, params.exp_name, params.exp_id, hyp_bin)
torch.save(generated_ks[0], hyp_bin_path, pickle_protocol=4)
print("binarized file save in %s" % hyp_bin_path)
'''
# write and eval_bleu
hyp_paths = []
for i in range(K):
hyp_name = "{0}_hyp_{1}-{2}.K{3}.txt".format(nist, params.src_lang, params.tgt_lang, i)
hyp_path = os.path.join(params.dump_path, params.exp_name, params.exp_id, hyp_name)
hyp_paths.append(hyp_path)
# export sentences to hypothesis file / restore BPE segmentation
with open(hyp_path, 'w', encoding='utf-8') as f:
f.write('\n'.join(hypothesis_ks[i]) + '\n')
restore_segmentation(hyp_path)
bleus = eval_mbleu(['%s/%s.tok.ref0'%(data_root, nist), '%s/%s.tok.ref1'%(data_root, nist), '%s/%s.tok.ref2'%(data_root, nist), '%s/%s.tok.ref3'%(data_root, nist)], hyp_paths)
logger.info("%s mBLEU (max): %f\n" % (nist, max(bleus)))
logger.info("%s mBLEU (avg): %f\n" % (nist, sum(bleus) / len(bleus)))
for i, b in enumerate(bleus):
logger.info("%s top %i confidence/ BLEU: %f/ %f\n" % (nist, i+1, mean_conf[i], b))
div_bleu = eval_diverse_bleu(hyp_paths)
logger.info("%s div_BLEU: %f\n" % (nist, div_bleu))
if __name__ == '__main__':
# generate parser / parse parameters
parser = get_parser()
params = parser.parse_args()
# check parameters
assert os.path.isfile(params.model_path)
assert os.path.isdir(params.dump_path)
assert params.src_lang != '' and params.tgt_lang != '' and params.src_lang != params.tgt_lang
# translate
with torch.no_grad():
main(params)