-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
212 lines (171 loc) · 6.68 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
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
"""**Train**.
This module train the News Recommendation System.
.. _Google Python Style Guide:
http://google.github.io/styleguide/pyguide.html
"""
import bcolz
import torch
from torch.utils import data
import pytorch_lightning as pl
from ranger import Ranger
from dataset import NewsDataset
from model.net import NRMS
from metric import ndcg_score, mrr_score
from config import hparams
from vocab import extract_glove_vocab
class Model(pl.LightningModule):
"""
Define how to train the model using LightningModule.
"""
def __init__(self, hparams):
"""Initialization of parameters.
Args:
params (dict): Dictionary of configuration parameters.
"""
super(Model, self).__init__()
self.hparams = hparams
self.embeddings = self.init_embedding()
self.model = NRMS(hparams['model'], self.embeddings)
def init_embedding(self):
"""Load pre-trained embeddings as a constant tensor.
Args:
file_path (str): the pre-trained embeddings filename.
Returns:
obj: A constant tensor.
"""
glove_path = self.hparams['glove_path']
embed_size = self.hparams['model']['embed_size']
max_vocab_size = self.hparams['max_vocab_size']
extract_glove_vocab(glove_path, embed_size, max_vocab_size)
vectors = bcolz.open(f'{glove_path}/6B.'+str(embed_size)+'.dat')[:]
embeddings = torch.Tensor(vectors)
if hparams['model']['dct_size'] == 'auto':
hparams['model']['dct_size'] = embeddings.shape[0]
return embeddings
def configure_optimizers(self):
"""Optimizer configuration.
Returns:
object: Optimizer.
"""
optimizer = Ranger( self.parameters(),
lr=self.hparams['lr'],
weight_decay=1e-5)
return optimizer
def setup(self,stage):
"""
Data set definition according to stage.
Args:
stage (str): Modeling Stage.
"""
if stage == 'fit':
train_ds = NewsDataset(self.hparams, self.hparams['path_train_data'])
val_ds = NewsDataset(self.hparams, self.hparams['path_val_data'])
self.train_ds, _ = data.random_split(train_ds, [len(train_ds)-int(len(train_ds)*0.99), int(len(train_ds)*0.99)])
self.val_ds, _ = data.random_split(val_ds, [len(val_ds)-int(len(val_ds)*0.95), int(len(val_ds)*0.95)])
if stage == 'test':
self.test_ds = NewsDataset(self.hparams, self.hparams['path_val_data'])
def train_dataloader(self):
"""Build Data loader from train dataset.
Returns:
dataloader: Train data loader.
"""
train_dataloader = data.DataLoader(self.train_ds,
num_workers=self.hparams['num_workers'],
batch_size=self.hparams['batch_size'],
shuffle=self.hparams['shuffle'])
return train_dataloader
def val_dataloader(self):
"""Build Data loader from validation dataset.
Returns:
dataloader: Validation data loader.
"""
val_dataloader = data.DataLoader(self.val_ds,
num_workers=self.hparams['num_workers'],
batch_size=self.hparams['batch_size'],
shuffle=self.hparams['shuffle'])
return val_dataloader
def test_dataloader(self):
"""Build Data loader from test dataset.
Returns:
dataloader: Test data loader.
"""
test_dataloader = data.DataLoader(self.test_ds,
num_workers=self.hparams['num_workers'],
batch_size=self.hparams['batch_size'],
shuffle=self.hparams['shuffle'])
return test_dataloader
def forward(self):
"""Forward.
Define as normal pytorch model.
"""
return None
def training_step(self, batch, _):
"""For each step(batch).
Args:
batch {[type]} -- data
batch_idx {[type]}
"""
clicks, cands, labels = batch
loss, _ = self.model(clicks, cands, labels)
return {'loss': loss}
def training_epoch_end(self, outputs):
"""For each epoch end.
Args:
outputs: Loss values.
Returns:
dict: Logs loss mean.
"""
loss_mean = torch.stack([x['loss'] for x in outputs]).mean()
logs = {'train_loss': loss_mean}
self.model.eval()
return {'progress_bar': logs, 'log': logs}
def validation_step(self, batch, _):
"""For each step(batch).
Args:
batch {[type]} -- data
batch_idx {[type]}
Returns:
dict: Evaluation metrics on training step.
"""
clicks, cands, cands_label = batch
with torch.no_grad():
logits = self.model(clicks, cands)
mrr = 0.
auc = 0.
ndcg5, ndcg10 = 0., 0.
for score, label in zip(logits, cands_label):
auc += pl.metrics.functional.auroc(score, label)
score = score.detach().cpu().numpy()
label = label.detach().cpu().numpy()
mrr += mrr_score(label, score)
ndcg5 += ndcg_score(label, score, 5)
ndcg10 += ndcg_score(label, score, 10)
auroc = (auc / logits.shape[0]).item()
mrr = (mrr / logits.shape[0]).item()
ndcg5 = (ndcg5 / logits.shape[0]).item()
ndcg10 = (ndcg10 / logits.shape[0]).item()
return {'auroc': auroc, 'mrr': mrr, 'ndcg5': ndcg5, 'ndcg10': ndcg10}
def validation_epoch_end(self, outputs):
"""Validation end.
Args:
outputs (dict): History per evaluation metric.
Reruns:
dict: Logs of metrics.
"""
mrr = torch.Tensor([x['mrr'] for x in outputs])
auroc = torch.Tensor([x['auroc'] for x in outputs])
ndcg5 = torch.Tensor([x['ndcg5'] for x in outputs])
ndcg10 = torch.Tensor([x['ndcg10'] for x in outputs])
logs = {'auroc': auroc.mean(),
'mrr': mrr.mean(),
'ndcg@5': ndcg5.mean(),
'ndcg@10': ndcg10.mean()}
self.model.train()
return {'progress_bar': logs, 'log': logs}
if __name__ == '__main__':
from pytorch_lightning import Trainer
from pytorch_lightning.profiler import AdvancedProfiler
profiler = AdvancedProfiler()
model = Model(hparams)
trainer = Trainer(max_epochs=2, profiler=profiler)
trainer.fit(model)