|
1
|
|
|
#!/usr/bin/env python |
|
2
|
|
|
# -*- coding: utf-8 -*- |
|
3
|
|
|
|
|
4
|
|
|
import logging as loggers |
|
5
|
|
|
from abc import ABCMeta, abstractmethod |
|
6
|
|
|
|
|
7
|
|
|
from deepy.trainers import NeuralTrainer |
|
8
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
logging = loggers.getLogger(__name__) |
|
11
|
|
|
THEANO_LINKER = 'cvm' |
|
12
|
|
|
|
|
13
|
|
|
class CustomizeTrainer(NeuralTrainer): |
|
14
|
|
|
'''This is a base class for all trainers.''' |
|
15
|
|
|
__metaclass__ = ABCMeta |
|
16
|
|
|
|
|
17
|
|
|
def __init__(self, network, config=None): |
|
18
|
|
|
""" |
|
19
|
|
|
Basic neural network trainer. |
|
20
|
|
|
:type network: deepy.NeuralNetwork |
|
21
|
|
|
:type config: deepy.conf.TrainerConfig |
|
22
|
|
|
:return: |
|
23
|
|
|
""" |
|
24
|
|
|
super(CustomizeTrainer, self).__init__(network, config) |
|
25
|
|
|
|
|
26
|
|
|
|
|
27
|
|
|
def train(self, train_set, valid_set=None, test_set=None, train_size=None): |
|
28
|
|
|
'''We train over mini-batches and evaluate periodically.''' |
|
29
|
|
|
iteration = 0 |
|
30
|
|
|
while True: |
|
31
|
|
|
if not iteration % self.config.test_frequency and test_set: |
|
32
|
|
|
try: |
|
33
|
|
|
self.test(iteration, test_set) |
|
34
|
|
|
except KeyboardInterrupt: |
|
35
|
|
|
logging.info('interrupted!') |
|
36
|
|
|
break |
|
37
|
|
|
|
|
38
|
|
|
if not iteration % self.validation_frequency and valid_set: |
|
39
|
|
|
try: |
|
40
|
|
|
if not self.evaluate(iteration, valid_set): |
|
41
|
|
|
logging.info('patience elapsed, bailing out') |
|
42
|
|
|
break |
|
43
|
|
|
except KeyboardInterrupt: |
|
44
|
|
|
logging.info('interrupted!') |
|
45
|
|
|
break |
|
46
|
|
|
|
|
47
|
|
|
train_message = "" |
|
48
|
|
|
try: |
|
49
|
|
|
train_message = self.train_func(train_set) |
|
50
|
|
|
except KeyboardInterrupt: |
|
51
|
|
|
logging.info('interrupted!') |
|
52
|
|
|
break |
|
53
|
|
|
if not iteration % self.config.monitor_frequency: |
|
54
|
|
|
logging.info('monitor (iter=%i) %s', iteration + 1, train_message) |
|
55
|
|
|
|
|
56
|
|
|
iteration += 1 |
|
57
|
|
|
if hasattr(self.network, "iteration_callback"): |
|
58
|
|
|
self.network.iteration_callback() |
|
59
|
|
|
|
|
60
|
|
|
yield train_message |
|
61
|
|
|
|
|
62
|
|
|
if valid_set: |
|
63
|
|
|
self.set_params(self.best_params) |
|
64
|
|
|
if test_set: |
|
65
|
|
|
self.test(0, test_set) |
|
66
|
|
|
|
|
67
|
|
|
@abstractmethod |
|
68
|
|
|
def train_func(self, train_set): |
|
69
|
|
|
return "" |
|
70
|
|
|
|
uaca /
deepy
