-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathqr_cnn.py
115 lines (95 loc) · 4.17 KB
/
qr_cnn.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
from datetime import datetime
import time
from src import cnn
from src.common import *
torch.cuda.set_device(0)
embeddings_p200 = EmbeddingsP200("./data_local/vectors_pruned.200.txt")
corpus = Corpus("./data_local/text_tokenized.txt", max_text_length=100)
train_data = TrainingData("./data_local/train_random.txt")
dev_data = read_android_annotations('./data_local/dev.txt')
dev_title_emb_list, dev_text_emb_list = batch_to_emb(dev_data, corpus=corpus, n_neg_samples=20, embeddings=embeddings_p200, train=False)
save_to_file = False
n_neg_samples = 20
hidden_dim = 667
batch_size = 10
batch_i = 0
tot_N = len(train_data.train_items)
n_batches = int(tot_N/batch_size)+1
n_cand = n_neg_samples+2
n_cand_dev = 21
emb_size = 200
kernel_size = 3
model = cnn.CNNQAModel(hidden_dim=hidden_dim, kernel_size=kernel_size, cand_size=n_cand, emb_size=emb_size, cand_size_dev=n_cand_dev, cuda=True)
max_margin_loss = torch.nn.MultiMarginLoss(margin = 0.3)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
now = datetime.now()
session_id = "{}{:02d}{:02d}_{:02d}{:02d}_".format(now.year, now.month, now.day, now.hour, now.minute)+str(int(time.time()))
result_folder = "results_cnn/"+ session_id
output_file = os.path.join(result_folder, "output.txt")
printf = get_printf(save_to_file=save_to_file, output_file=output_file)
for n_epoch in range(6):
train_data.shuffle()
printf("Cur Epoch:", n_epoch)
for batch_i in range(n_batches):
optimizer.zero_grad()
cur_batch = train_data.train_items[batch_size*batch_i : min(batch_size*batch_i+batch_size, tot_N)]
# print(batch_i, n_batches)
cur_batch_size = len(cur_batch)
title_emb_list, text_emb_list = batch_to_emb(cur_batch, corpus=corpus, n_neg_samples=n_neg_samples, embeddings=embeddings_p200)
text_emb, text_lengths, text_rev = pad_and_sort(text_emb_list)
title_emb, title_lengths, title_rev = pad_and_sort(title_emb_list)
text_rev = cuda(text_rev)
title_rev = cuda(title_rev)
title_emb_var = to_variable(title_emb)
text_emb_var = to_variable(text_emb)
title_length_var = to_variable(torch.from_numpy(title_lengths).float())
text_length_var = to_variable(torch.from_numpy(text_lengths).float())
cos_sim = model(title_emb_var, title_length_var, title_rev, text_emb_var, text_length_var, text_rev, cur_batch_size)
const_y = to_variable(torch.from_numpy(np.array([0]*len(cur_batch))))
# const_y = torch.autograd.Variable(torch.from_numpy(np.array([1]*len(cur_batch))), requires_grad=False)
loss = max_margin_loss(cos_sim, const_y)
loss.backward()
optimizer.step()
if (batch_i+1) % 50 == 0:
printf ('Epoch %d batch %d, Loss: %.4f' % (n_epoch+1, batch_i+1, loss.data[0]))
if batch_i == n_batches-1:
# Dev data, feed into model 1 by 1
dev_pred = [ ]
for dev_i in range(len(dev_title_emb_list)):
title_emb = dev_title_emb_list[dev_i]
text_emb = dev_text_emb_list[dev_i]
text_emb, text_lengths, text_rev = pad_and_sort([text_emb])
title_emb, title_lengths, title_rev = pad_and_sort([title_emb])
text_rev = cuda(text_rev)
title_rev = cuda(title_rev)
title_emb_var = to_variable(title_emb)
text_emb_var = to_variable(text_emb)
title_length_var = to_variable(torch.from_numpy(title_lengths).float())
text_length_var = to_variable(torch.from_numpy(text_lengths).float())
cos_sim = model(title_emb_var, title_length_var, title_rev, text_emb_var, text_length_var, text_rev, 1, dev=True)
cos_arr = cos_sim.cpu().data.numpy()[0]
true_labels = dev_data[dev_i][2]
pred_tmp = [x for _,x in sorted(zip(cos_arr, true_labels), reverse=True)]
dev_pred.append(pred_tmp)
metric = Evaluation(dev_pred)
printf('Dev metric:')
printf('MAP:', metric.MAP())
printf('MRR:', metric.MRR())
printf('P@1:', metric.Precision(1))
printf('P@5:', metric.Precision(5))
# print('Dev done')
del cos_sim
del const_y
del title_emb_var
del text_emb_var
del title_length_var
del text_length_var
del text_emb, text_lengths, text_rev
del title_emb, title_lengths, title_rev
torch.cuda.empty_cache()
fn = os.path.join(result_folder, "model_epoch_{}".format(n_epoch+1))
if save_to_file:
torch.save({
"model": model.state_dict(),
"opt": optimizer.state_dict()
}, fn+"_model")