-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathpredict.py
193 lines (162 loc) · 7.58 KB
/
predict.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
# -*- coding: utf-8 -*-
from __future__ import unicode_literals, print_function, division
import torch
import time
import os
import json
import argparse
import random
import importlib
import configparser
import torch.utils.data as Data
import data_utils
import constrains
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print('device:', device)
# seed = 1
# torch.manual_seed(seed)
# torch.cuda.manual_seed_all(seed)
# np.random.seed(seed)
# random.seed(seed)
# torch.backends.cudnn.deterministic = True
# word dict
with open('resource/word_dict.json', 'r', encoding='utf-8') as f1:
word_dict = json.load(f1)
def predict(test_data, lines, targets, model, predict_param, ckpt_path, poem_type, cangtou):
model.eval()
context = ''
for i, data in enumerate(test_data):
# 每一步 loader 释放一小批数据用来学习，step=总数据量/batch_size，enumerate把每次提取编写索引。
# batch_x: B*T tensor
output_words = model.predict(data, cangtou, predict_param)
output_words.insert(7, '/') # 改
output_words.insert(15, '/')
output_words.insert(23, '/')
output_sentence = ''.join(output_words)
context_poem = lines[i] + ' ==== ' + output_sentence + '\n'
# context_poem = str(i + 1) + '\n' + lines[i] + ' ==== ' + output_sentence + '\n' # Jul3
context = context + context_poem
if i < 3:
print(context_poem)
if targets:
target = lines[i] + ' ==== ' + targets[i] + '\n'
# context = context + target # Jul3
if i < 3:
print(target)
# logs
if len(ckpt_path.split('/'))==2:
file = 'result/' + ckpt_path.split('/')[1].split('.pkl')[0] + '_' + str(time.strftime("%H%M%S", time.localtime())) + '.txt'
else:
file = 'result/' + ckpt_path.split('/')[2].split('.pkl')[0] + '_' + str(time.strftime("%H%M%S", time.localtime())) + '.txt'
with open(file, 'w', encoding='utf-8') as f:
f.write(context)
print('result saved at:', file)
return file
def pred(predict_param):
# ========= Loading Params =========#
# predict param
parser = argparse.ArgumentParser(description='Vivi')
parser.add_argument('--model_name', type=str, default='Seq2seq_12')
# parser.add_argument('--ckpt_path', type=str, default='ckpt/0802164334_Seq2seq_12_ep=3_loss=150.38.pkl')
# parser.add_argument('--ckpt_path', type=str, default='ckpt/0802164334_Seq2seq_12_ep=4_loss=149.02.pkl')
parser.add_argument('--ckpt_path', type=str, default='ckpt/0802164555_Seq2seq_12_ep=3_loss=154.49.pkl')
# parser.add_argument('--ckpt_path', type=str, default='ckpt/0802164555_Seq2seq_12_ep=4_loss=152.79.pkl')
# parser.add_argument('--ckpt_path', type=str, default='ckpt/0802164555_Seq2seq_12_ep=5_loss=151.01.pkl')
# parser.add_argument('--ckpt_path', type=str, default='ckpt/0802164555_Seq2seq_12/0802164555_Seq2seq_12_ep=6_loss=150.18.pkl')
parser.add_argument('--cangtou', type=str, default='')
parser.add_argument('--keywords', type=str, default='')
parser.add_argument('--test_set', type=str, default='')
parser.add_argument('--eval_set', type=str, default='resource/dataset/poem_1031k_theme_test_1k.txt')
parser.add_argument('--use_planning', type=bool, default=False)
parser.add_argument('--bleu_eval', type=bool, default=False)
parser.add_argument('--poem_type', type=str, default='poem7')
parser.add_argument('--train_mode', type=str, default='kw2poem') # nL21L or kw2poem
parser.add_argument('--note', type=str, default='')
parser.add_argument('--as_train', type=bool, default=False) # Jul23
parser.add_argument('--pred_soft', type=bool, default=True)
parser.add_argument('--template', type=bool, default=False) # 2 template T, 4 soft F
parser.add_argument('--hard_rhyme', type=bool, default=True)
parser.add_argument('--hard_tone', type=bool, default=True)
parser.add_argument('--w1', type=float, default=4.)
parser.add_argument('--w2', type=float, default=0.)
args = parser.parse_args()
if predict_param == {}:
print('pred param from args')
predict_param = vars(args)
else:
print('pred param from py script')
model_name = predict_param['model_name']
ckpt_path = predict_param['ckpt_path']
cangtou = predict_param['cangtou']
keywords = predict_param['keywords']
test_set = predict_param['test_set']
eval_set = predict_param['eval_set']
use_planning = predict_param['use_planning']
poem_type = predict_param['poem_type']
train_mode = predict_param['train_mode']
as_train = predict_param['as_train']
# model param
conf = configparser.ConfigParser()
conf.read('config/config_'+model_name+'.ini')
model_param_li = conf.items('model_param')
model_param = {}
for item in model_param_li:
model_param[item[0]] = item[1]
# ========= Preparing Data =========#
# read data
targets = None
if cangtou:
test_set, lines = data_utils.get_keywords(cangtou, use_planning)
elif keywords:
test_set, lines = data_utils.get_keywords(keywords, use_planning)
elif test_set:
test_set, lines = data_utils.read_test_data(test_set, use_planning)
elif train_mode == 'nL21L':
test_set, lines, targets = data_utils.read_nL21L_eval_data(eval_set)
else: # eval
test_set, lines, targets = data_utils.read_eval_data(eval_set, use_planning) # read_eval_data May24
# 实例化
data_path = 'models.' + model_name + '.PoetryData'
PoetryData = importlib.import_module(data_path)
test_Dataset = getattr(PoetryData, 'PoetryData')(test_set, src_max_len=int(model_param['input_max_len']),
tgt_max_len=int(model_param['target_max_len']), test=True)
# 变成小批
test_data = Data.DataLoader(
dataset=test_Dataset, # torch TensorDataset format
batch_size=1,
shuffle=False,
collate_fn=PoetryData.collate_fn # Jun16
# num_workers=2, # 多线程来读数据，提取xy的时候几个数据一起提取
)
# ========= Preparing Model =========#
if os.path.exists(ckpt_path):
checkpoint = torch.load(ckpt_path, map_location=lambda storage, loc: storage)
# model_param = checkpoint['settings'] # Jun16
model_param = checkpoint['model_param']
model_path = 'models.' + model_name + '.' + model_name
Model = importlib.import_module(model_path) # 导入模块
model = getattr(Model, model_name)(model_param) # 反射并实例化
print('model:', model)
model.load_state_dict(checkpoint['model'])
if 'train_param' in checkpoint.keys():
print('train param:', checkpoint['train_param'])
# Jul23
if as_train:
train_param = checkpoint['train_param']
predict_param['ckpt_path'] = ckpt_path
predict_param['pred_soft'] = train_param['train_soft']
predict_param['template'] = train_param['template']
predict_param['w1'] = train_param['w1']
predict_param['w2'] = train_param['w2']
print('predict param: ', predict_param)
else:
print('train param not recorded.')
save_file = predict(test_data, lines, targets, model, predict_param, ckpt_path, poem_type, cangtou)
else:
print('ckpt_path does not exist.')
save_file = None
return save_file
def main():
save_file = pred({})
if __name__ == '__main__':
main()