-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrainer.py
238 lines (205 loc) · 10 KB
/
trainer.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
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
import argparse
import json
import os
import sys
from argparse import ArgumentParser
import pytorch_lightning as pl
import time
from callbacks.logParameters import LogParameters
from callbacks.lossCurveCallback import LossCurveCallback
from loggers.log import Log_and_print
from models.qAModel import VQAModelClassifier
from utils import collect_env_details
lstr_args = ['--max_epochs','3']
def cli_main(parser):
print('MAIN START')
ts_script = time.time()
# ------------
# args
# ------------
print('MAIN args')
# program level args
parser.add_argument('--seed', default=0, type=int)
parser.add_argument('--save_dir', default='./output/results', type=str)
parser.add_argument('--run_name', default='default_run', type=str)
parser.add_argument('--log_parameters', default=0, type=int)
parser.add_argument('--es_patience', default=20, type=int)
parser.add_argument('--figsize_x', default=15, type=float)
parser.add_argument('--figsize_y', default=10, type=float)
# trainer level args
parser = pl.Trainer.add_argparse_args(parser)
# model level args
parser = VQAModelClassifier.add_model_specific_args(parser)
args = parser.parse_args(lstr_args)
# always print full weights_summary
args.weights_summary = 'full'
# automatically use all available GPUs
# https://pytorch-lightning.readthedocs.io/en/latest/advanced/multi_gpu.html#select-gpu-devices
args.gpus = -1
# seed
pl.seed_everything(args.seed)
# logger
# https://pytorch-lightning.readthedocs.io/en/latest/common/loggers.html#tensorboard
# https://pytorch-lightning.readthedocs.io/en/latest/extensions/logging.html
# loggers need info from args, so have to run args first before loggers
tb_logger = pl.loggers.TensorBoardLogger(save_dir = args.save_dir+'log/',
name = args.run_name,
version = 'fixed_version',
log_graph = True)
'''
The tensorboard is creating a new version unless we fix it with a new version name.
'''
wandb_logger = pl.loggers.WandbLogger(save_dir = args.save_dir+'log/',
offline = True, # cannot log model while offline
name = args.run_name,
version = 'fixed_version')
lnp = Log_and_print(tb_logger, wandb_logger, args.run_name)
lnp.lnp('Loggers start')
lnp.lnp('ts_script: ' + str(ts_script))
sys.path += [os.path.abspath(".."), os.path.abspath(".")]
lnp.lnp(collect_env_details())
strargs = ''
for (k,v) in vars(args).items():
strargs += str(k) + ': ' + str(v) + '\n'
lnp.lnp('ARGUMENTS:\n' + strargs)
# ------------
# LightningModule
# https://pytorch-lightning.readthedocs.io/en/latest/common/lightning_module.html
# ------------
lnp.lnp('MAIN LightningModule')
lm = VQAModelClassifier(**vars(args))
for n,p in lm.named_parameters():
lnp.lnp(n + ': ' + str(p.data.shape))
# ------------
# training
# https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html
# ------------
# Callbacks
# https://pytorch-lightning.readthedocs.io/en/latest/extensions/callbacks.html
lnp.lnp('MAIN callbacks')
l_callbacks = []
# custom loss curve
cbLossCurveCallback = LossCurveCallback(lnp, figsize=[args.figsize_x, args.figsize_y]) # custom
l_callbacks.append(cbLossCurveCallback)
# early stopping
# https://pytorch-lightning.readthedocs.io/en/latest/common/early_stopping.html
#cbEarlyStopping = pl.callbacks.early_stopping.EarlyStopping(monitor='val_loss', patience=args.es_patience)
#l_callbacks.append(cbEarlyStopping)
# model checkpoint
# https://pytorch-lightning.readthedocs.io/en/latest/common/weights_loading.html#automatic-saving
checkpoint_dirpath = args.save_dir + 'checkpoints/'
checkpoint_filename = args.save_dir[:-1] + '_' + args.run_name
lnp.lnp('checkpoint_dirpath: ' + checkpoint_dirpath)
lnp.lnp('checkpoint_filename: ' + checkpoint_filename)
cbModelCheckpoint = pl.callbacks.ModelCheckpoint(monitor='val_loss', dirpath=checkpoint_dirpath, filename=checkpoint_filename)
l_callbacks.append(cbModelCheckpoint)
# log parameters
if args.log_parameters:
cbLogParameters = LogParameters()
l_callbacks.append(cbLogParameters)
lnp.lnp('MAIN trainer')
trainer = pl.Trainer.from_argparse_args(args,
logger=[tb_logger, wandb_logger],
callbacks=l_callbacks,
)
trainer.val_percent_check = 0
trainer.check_val_every_n_epoch = 3
# LEARNING RATE FINDER
# https://pytorch-lightning.readthedocs.io/en/latest/advanced/lr_finder.html#learning-rate-finder
# MisconfigurationException: No `train_dataloader()` method defined. Lightning `Trainer` expects as minimum a `training_step()`, `train_dataloader()` and `configure_optimizers()` to be defined.
"""lr_finder = trainer.tuner.lr_find(lm)
fig = lr_finder.plot(suggest=True)
trainer.logger[0].experiment.add_figure('lr_finder', fig)
trainer.logger[1].experiment.log({'lr_finder': fig})
# TODO: log to wandb too https://docs.wandb.ai/guides/track/log/plots#matplotlib-and-plotly-plots
new_lr = lr_finder.suggestion()
if new_lr is None:
lnp.lnp('new_lr was not found. Using default lr: ' + str(lm.hparams.lr))
else:
lnp.lnp('new_lr: ' + str(new_lr))
lm.hparams.lr = new_lr"""
# fit
lnp.lnp('MAIN fit')
"""train_dataload = train_dataloader()
val_dataload = val_dataloader()
trainer.fit(lm, train_dataload, val_dataload)"""
trainer.fit(lm)
# ------------
# testing
# ------------
lnp.lnp('MAIN test')
ts_test = time.time()
# test from the best checkpoint
# https://pytorch-lightning.readthedocs.io/en/latest/common/test_set.html
test_output = trainer.test(ckpt_path = 'best')
tf_test = time.time()
lnp.lnp('test_output: ' + str(test_output))
lnp.lnp('ts_test: ' + str(ts_test))
lnp.lnp('tf_test: ' + str(tf_test))
dur_test = tf_test - ts_test
lnp.lnp('Test duration: ' + str(dur_test))
# JSON log
dlog = {'lowest_val' : float(cbLossCurveCallback.a_val_loss.min()),
'a_lowest_val' : int(cbLossCurveCallback.a_val_loss.argmin()),
'current_epoch' : trainer.current_epoch,
'dur_test' : dur_test,
'tst_loss' : test_output[0]['tst_loss'],
'new_lr' : lm.hparams.lr,
'n_trainable_params' : sum(p.numel() for p in lm.parameters() if p.requires_grad)}
dlog.update(vars(args))
tf_script = time.time()
lnp.lnp('tf_script: ' + str(tf_script))
dur_script = tf_script - ts_script
lnp.lnp('Script duration: ' + str(dur_script))
dlog['dur_script'] = dur_script
# Tear down
lnp.lnp('MAIN END (only logging is left)')
with open(args.save_dir + '/results' + '_' + args.run_name +'.json', "w") as outfile:
json.dump(dlog, outfile)
lnp.dump_to_tensorboard()
lnp.dump_to_wandb()
print('everything done')
return trainer
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Full Pipeline Training')
parser.add_argument('--train_batch_size', type=int, default=8,
help='Shorter side transformation.')
parser.add_argument('--eval_batch_size', type=int, default=8,
help='Shorter side transformation.')
parser.add_argument('--num_workers', type=int, default=1,
help='Shorter side transformation.')
parser.add_argument('--n_val', type=int, default=5000,
help='Shorter side transformation.')
parser.add_argument('--n_train', type=int, default=10000,
help='Shorter side transformation.')
parser.add_argument('--n_test', type=int, default=5000,
help='Shorter side transformation.')
parser.add_argument('--train_data_dir', type=str, default="./data",
help='Shorter side transformation.')
parser.add_argument('--val_data_dir', type=str, default="./data",
help='Shorter side transformation.')
parser.add_argument('--test_data_dir', type=str, default="./data",
help='Shorter side transformation.')
parser.add_argument('--train_answersDataSubType', type=str, default="train2014",
help='Shorter side transformation.')
parser.add_argument('--val_answersDataSubType', type=str, default="val2014",
help='Shorter side transformation.')
parser.add_argument('--test_answersDataSubType', type=str, default="val2014",
help='Shorter side transformation.')
parser.add_argument('--train_questionDataSubType', type=str, default="train2014",
help='Shorter side transformation.')
parser.add_argument('--val_questionDataSubType', type=str, default="val2014",
help='Shorter side transformation.')
parser.add_argument('--test_questionDataSubType', type=str, default="val2014",
help='Shorter side transformation.')
parser.add_argument('--numCandidates', type=int, default=5,
help='Shorter side transformation.')
parser.add_argument('--trainPklFilePath', type=str, default='./output/intermediate/trainNormalisedFeatures.pkl',
help='Shorter side transformation.')
parser.add_argument('--valPklFilePath', type=str, default='./output/intermediate/normalisedFeatures.pkl',
help='Shorter side transformation.')
parser.add_argument('--resultsTrain', type=str, default='resultsTrain',
help='Shorter side transformation.')
parser.add_argument('--resultsVal', type=str, default='resultsVal',
help='Shorter side transformation.')
trainer = cli_main(parser)