-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
182 lines (128 loc) · 5.27 KB
/
train.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
import torch
from torch.autograd import Variable
from utils import data_loader, field_size
from timeit import default_timer
from torch_utils.utils import logging
def train(model, logger, data_index=1458, _type='IPNN', embedding_method=None, refresh=False, minibatch=200, device_ids=[0]):
logger.update("data_index: {}".format(data_index), key='define')
_loader, column_path = data_loader(data_index=data_index, refresh=refresh, minibatch=minibatch, shuffle=True, gpu=True)
fields = field_size(column_path)
input_dims = sum(fields)+1
if(_type=='IPNN'):
logger.update("method: IPNN", key='define')
logger.update("embedding_method to be: {}".format(embedding_method.__name__), key='define')
model = model(fields, embedding_method=embedding_method, embedding_dim=10).cuda()
elif(_type=='LR'):
logger.update("method: LR", key='define')
model = model(input_dims).cuda()
# model = torch.nn.DataParallel(model, device_ids=device_ids)
loss = torch.nn.NLLLoss().cuda()
lr = 1e-3
solver = torch.optim.Adam(model.parameters(), lr=lr)
for _iter in range(1000):
end = default_timer()
_loader.create_iter()
LOSS, sub_accumulate_correct_pre, sub_accumulate_correct_after = 0, 0, 0
_sum = {}
_sum_pre = {}
index = 1
for tensor, _labels in _loader._iter:
if(embedding_method.__name__=='linear_field'):
tensor = tensor.float()
else:
tensor = tensor.long()
for inner_iter in range(1000):
inputs = Variable(tensor)
labels = Variable(_labels)
outputs = model(inputs)
_loss = loss(outputs, labels)
LOSS = LOSS+_loss.data[0]
_loss.backward()
solver.step()
if(inner_iter==0):
pred_pre = outputs.data.max(1)[1]
pred = outputs.data.max(1)[1]
correct_pre = pred.eq(labels.data).sum()
sub_accumulate_correct_pre += pred.eq(labels.data).sum()
if(inner_iter==999):
pred = outputs.data.max(1)[1]
correct_after = pred.eq(labels.data).sum()
sub_accumulate_correct_after += pred.eq(labels.data).sum()
for x in range(pred.shape[0]):
if pred[x] not in list(_sum.keys()):
_sum[pred[x]] = 1
else:
_sum[pred[x]] = _sum[pred[x]]+1
for x in range(pred_pre.shape[0]):
if pred_pre[x] not in list(_sum_pre.keys()):
_sum_pre[pred_pre[x]] = 1
else:
_sum_pre[pred_pre[x]] = _sum_pre[pred_pre[x]]+1
time = default_timer()-end
end = default_timer()
string_pre = "index: {}, time: {}, sub_dict: {}, correct: {}, sub_accumulate_correct: {}".format(index, time, _sum_pre, correct_pre*1./(labels.data.shape[0]), sub_accumulate_correct_pre*1./(index*labels.data.shape[0]) )
string_after = "index: {}, time: {}, sub_dict: {}, correct: {}, sub_accumulate_correct: {}".format(index, time, _sum, correct_after*1./(labels.data.shape[0]), sub_accumulate_correct_after*1./(index*labels.data.shape[0]) )
logger.update(string_pre, key='pre')
logger.update(string_after, key='after')
if(index==350):
break
index = index+1
break
correct = 100.*correct/_loader._len
LOSS = LOSS/_loader._len
print("loss: {:6f}, accuracy: {:6f}, len: {}, step_time: {:4f}".format(LOSS, correct, _loader._len, default_timer()-end))
print("... dict: {}".format(_sum))
def debug(model, minibatch=200, device_ids=[0]):
_loader, column_path = data_loader(debug=True, refresh=True, minibatch=minibatch, shuffle=False, gpu=True, truncate=True)
fields = field_size(column_path)
embedding_dim = 10
model = model(fields, embedding_dim).cuda()
# model = torch.nn.DataParallel(model, device_ids=device_ids)
loss = torch.nn.NLLLoss().cuda()
lr = 1e-3
solver = torch.optim.Adam(model.parameters(), lr=lr)
for _iter in range(10):
end = default_timer()
_loader.create_iter()
LOSS, correct = 0, 0
_sum = {}
for tensor, labels in _loader._iter:
tensor = tensor.long()
inputs = Variable(tensor)
labels = Variable(labels)
outputs = model(inputs)
_loss = loss(outputs, labels)
LOSS = LOSS+_loss.data[0]
_loss.backward()
solver.step()
pred = outputs.data.max(1)[1]
correct += pred.eq(labels.data).sum()
for x in range(pred.shape[0]):
if pred[x] not in list(_sum.keys()):
_sum[pred[x]] = 1
else:
_sum[pred[x]] = _sum[pred[x]]+1
break
correct = 100.*correct/_loader._len
LOSS = LOSS/_loader._len
print("loss: {:6f}, accuracy: {:6f}, len: {}, step_time: {:4f}".format(LOSS, correct, _loader._len, default_timer()-end))
print("... dict: {}".format(_sum))
# _loader.create_iter()
# for tensor, labels in _loader._iter:
# tensor = tensor.long()
# print(tensor.shape, labels.shape, tensor.type(), labels.type())
# outputs = model(Variable(tensor))
# print(outputs.data.type(), outputs.data.shape)
# break
if __name__ == '__main__':
from IPNN import IPNN, linear_field, embedding_field
from LR import LR
# debug(IPNN, minibatch=20, device_ids=[0])
# train(IPNN, embedding_field, minibatch=200, device_ids=[0])
for index in [2821, 2997, 3358, 3386, 3427, 3476]:
targetDir = "./log/"
logger = logging(targetDir, onscreen=True)
train(IPNN, logger, data_index=index, _type='IPNN', refresh=True, embedding_method=linear_field, minibatch=200, device_ids=[0])
targetDir = "./log/"
logger = logging(targetDir, onscreen=True)
train(LR, logger, data_index=index, _type='LR', refresh=False, embedding_method=linear_field, minibatch=200, device_ids=[0])