/
score.py
144 lines (97 loc) · 4.47 KB
/
score.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
import torch
import torch.nn as nn
import argparse
import misc.utils as utils
import torch.utils.data as Data
from misc.HybridCNNLong import HybridCNNLong as DocumentCNN
from misc.LanguageModel import layer as LanguageModel
parser = utils.make_parser()
parser.add_argument('--start_from_file', type=int)
parser.add_argument('--end_to_file', type=int)
args = parser.parse_args()
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
from misc.dataloader import Dataloader
# get dataloader
data = Dataloader(args.input_json, args.input_ques_h5)
test_loader = Data.DataLoader(Data.Subset(data, range(args.train_dataset_len, args.train_dataset_len + args.val_dataset_len)), batch_size = args.batch_size, shuffle=True)
import math
decay_factor = math.exp(math.log(0.1) / (1500 * 1250))
import misc.net_utils as net_utils
from pycocoevalcap.eval import COCOEvalCap
def getObjsForScores(real_sents, pred_sents):
class coco:
def __init__(self, sents):
self.sents = sents
self.imgToAnns = [[{'caption' : sents[i]}] for i in range(len(sents))]
def getImgIds(self):
return [i for i in range(len(self.sents))]
return coco(real_sents), coco(pred_sents)
import time
import os
if args.start_from != 'None':
print('loading model from ' + args.start_from)
folder_name = args.start_from.split('/')[-1]
s = args.start_from_file
e = args.end_to_file
import subprocess
print(folder_name)
subprocess.run(['mkdir', os.path.join('result', folder_name)])
print(folder_name, "is created !!!")
for i in range(s, e+1):
print("Evaluating model number", i)
encoder = DocumentCNN(data.getVocabSize(), args.txtSize, dropout=args.drop_prob_lm, avg=1, cnn_dim=args.cnn_dim)
generator = LanguageModel(args.input_encoding_size, args.rnn_size, data.getSeqLength(), data.getVocabSize(), num_layers=args.rnn_layers, dropout=args.drop_prob_lm)
load_file = os.path.join(args.start_from, str(i) + '_-1.tar')
file_sample = os.path.join('result' ,folder_name, str(i))
file_score = file_sample + '-score'
checkpoint = torch.load(load_file, map_location=torch.device('cpu'))
encoder.load_state_dict(checkpoint['encoder_state_dict'])
generator.load_state_dict(checkpoint['generator_state_dict'])
encoder.eval()
generator.eval()
pred_sent = []
gt_sent = []
idx = 1
encoder = encoder.to(device)
generator = generator.to(device)
encoder = nn.DataParallel(encoder)
generator = nn.DataParallel(generator)
with torch.no_grad():
for input_sentence, lengths, sim_seq, _, _ in test_loader:
input_sentence = input_sentence.to(device)
lengths = lengths.to(device)
sim_seq = sim_seq.to(device)
encoded_input = encoder(utils.one_hot(input_sentence, data.getVocabSize()))
seq_logprob = generator(encoded_input, teacher_forcing=False)
seq_prob = torch.exp(seq_logprob)
seq = net_utils.prob2pred(seq_logprob)
# local loss criterion
loss = nn.CrossEntropyLoss(ignore_index=data.PAD_token)
# compute local loss
local_loss = loss(seq_logprob.permute(0, 2, 1), sim_seq)
# get encoding from
encoded_output = encoder(seq_prob)
encoded_sim = encoder(utils.one_hot(sim_seq, data.getVocabSize()))
# compute global loss
global_loss = net_utils.JointEmbeddingLoss(encoded_output, encoded_sim)
seq = seq.long()
sents = net_utils.decode_sequence(data.ix_to_word, seq)
real_sents = net_utils.decode_sequence(data.ix_to_word, input_sentence)
out_sents = net_utils.decode_sequence(data.ix_to_word, sim_seq)
pred_sent = pred_sent + sents
gt_sent = gt_sent + out_sents
f_sample = open(file_sample + '.txt', 'a')
for r, s, t in zip(real_sents,out_sents, sents):
f_sample.write(str(idx) + '\nreal : ' + r + '\nout : ' + s + '\npred : ' + t + '\n\n')
idx += 1
f_sample.close()
torch.cuda.empty_cache()
print('prediction completed...')
coco, cocoRes = getObjsForScores(gt_sent, pred_sent)
evalObj = COCOEvalCap(coco, cocoRes)
evalObj.evaluate()
f_score = open(file_score + '.txt', 'w')
for key in evalObj.eval:
f_score.write(key + ' : ' + str(evalObj.eval[key]) + '\n')
f_score.close()
print('Done !!!')