-
Notifications
You must be signed in to change notification settings - Fork 35
/
Copy pathsolver.py
159 lines (128 loc) · 6 KB
/
solver.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import os
import subprocess
from models import *
from utils import *
from parse import *
import random
from bert_optimizer import BertAdam
class Solver():
def __init__(self, args):
self.args = args
self.model_dir = make_save_dir(args.model_dir)
self.no_cuda = args.no_cuda
if not os.path.exists(os.path.join(self.model_dir,'code')):
os.makedirs(os.path.join(self.model_dir,'code'))
self.data_utils = data_utils(args)
self.model = self._make_model(self.data_utils.vocab_size, 10)
self.test_vecs = None
self.test_masked_lm_input = []
def _make_model(self, vocab_size, N=10,
d_model=512, d_ff=2048, h=8, dropout=0.1):
"Helper: Construct a model from hyperparameters."
c = copy.deepcopy
attn = MultiHeadedAttention(h, d_model, no_cuda=self.no_cuda)
group_attn = GroupAttention(d_model, no_cuda=self.no_cuda)
ff = PositionwiseFeedForward(d_model, d_ff, dropout)
position = PositionalEncoding(d_model, dropout)
word_embed = nn.Sequential(Embeddings(d_model, vocab_size), c(position))
model = Encoder(EncoderLayer(d_model, c(attn), c(ff), group_attn, dropout),
N, d_model, vocab_size, c(word_embed))
for p in model.parameters():
if p.dim() > 1:
nn.init.xavier_uniform(p)
if self.no_cuda:
return model
else:
return model.cuda()
def train(self):
if self.args.load:
path = os.path.join(self.model_dir, 'model.pth')
self.model.load_state_dict(torch.load(path)['state_dict'])
tt = 0
for name, param in self.model.named_parameters():
if param.requires_grad:
#print(name)
ttt = 1
for s in param.data.size():
ttt *= s
tt += ttt
print('total_param_num:',tt)
data_yielder = self.data_utils.train_data_yielder()
optim = torch.optim.Adam(self.model.parameters(), lr=1e-4, betas=(0.9, 0.98), eps=1e-9)
#optim = BertAdam(self.model.parameters(), lr=1e-4)
total_loss = []
start = time.time()
total_step_time = 0.
total_masked = 0.
total_token = 0.
for step in range(self.args.num_step):
self.model.train()
batch = data_yielder.__next__()
step_start = time.time()
out,break_probs = self.model.forward(batch['input'].long(), batch['input_mask'])
loss = self.model.masked_lm_loss(out, batch['target_vec'].long())
optim.zero_grad()
loss.backward()
torch.nn.utils.clip_grad_norm_(self.model.parameters(), 1.5)
optim.step()
total_loss.append(loss.detach().cpu().numpy())
total_step_time += time.time() - step_start
if step % 200 == 1:
elapsed = time.time() - start
print("Epoch Step: %d Loss: %f Total Time: %f Step Time: %f" %
(step, np.mean(total_loss), elapsed, total_step_time))
self.model.train()
print()
start = time.time()
total_loss = []
total_step_time = 0.
if step % 1000 == 0:
print('saving!!!!')
model_name = 'model.pth'
state = {'step': step, 'state_dict': self.model.state_dict()}
torch.save(state, os.path.join(self.model_dir, model_name))
def test(self, threshold=0.8):
path = os.path.join(self.model_dir, 'model.pth')
self.model.load_state_dict(torch.load(path)['state_dict'])
self.model.eval()
txts = get_test(self.args.test_path)
vecs = [self.data_utils.text2id(txt, 60) for txt in txts]
masks = [np.expand_dims(v != 0, -2).astype(np.int32) for v in vecs]
self.test_vecs = cc(vecs, no_cuda=self.no_cuda).long()
self.test_masks = cc(masks, no_cuda=self.no_cuda)
self.test_txts = txts
self.write_parse_tree()
def write_parse_tree(self, threshold=0.8):
batch_size = self.args.batch_size
result_dir = os.path.join(self.model_dir, 'result/')
make_save_dir(result_dir)
f_b = open(os.path.join(result_dir,'brackets.json'),'w')
f_t = open(os.path.join(result_dir,'tree.txt'),'w')
for b_id in range(int(len(self.test_txts)/batch_size)+1):
out,break_probs = self.model.forward(self.test_vecs[b_id*batch_size:(b_id+1)*batch_size],
self.test_masks[b_id*batch_size:(b_id+1)*batch_size])
for i in range(len(self.test_txts[b_id*batch_size:(b_id+1)*batch_size])):
length = len(self.test_txts[b_id*batch_size+i].strip().split())
bp = get_break_prob(break_probs[i])[:,1:length]
model_out = build_tree(bp, 9, 0, length-1, threshold)
if (0, length) in model_out:
model_out.remove((0, length))
if length < 2:
model_out = set()
f_b.write(json.dumps(list(model_out))+'\n')
"""
overlap = model_out.intersection(std_out)
prec = float(len(overlap)) / (len(model_out) + 1e-8)
reca = float(len(overlap)) / (len(std_out) + 1e-8)
if len(std_out) == 0:
reca = 1.
if len(model_out) == 0:
prec = 1.
f1 = 2 * prec * reca / (prec + reca + 1e-8)
"""
nltk_tree = dump_tree(bp, 9, 0, length-1, self.test_txts[b_id*batch_size+i].strip().split(), threshold)
f_t.write(str(nltk_tree).replace('\n','').replace(' ','') + '\n')