-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
161 lines (131 loc) · 8.41 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
#!/usr/bin/python3
# python3 train.py -i /home/jtan/gpu/jtan/data/pivot_Mbc_train.h5 -test /home/jtan/gpu/jtan/data/pivot_Mbc_test.h5 -n penalty_alt -lambda 10 -MI -JSD
import tensorflow as tf
import numpy as np
import pandas as pd
import time, os, sys
import argparse
# User-defined
from network import Network
from utils import Utils
from data import Data
from model import Model
from config import config_train, directories
tf.logging.set_verbosity(tf.logging.ERROR)
def train(config, args):
assert(config.use_adversary is False), 'To use adversarial training, run `adv_train.py`'
start_time = time.time()
global_step, n_checkpoints, v_auc_best, v_cvm = 0, 0, 0., 10.
ckpt = tf.train.get_checkpoint_state(directories.checkpoints)
print('Reading data ...')
if args.input is None:
input_file = directories.train
test_file = directories.test
else:
input_file = args.input
test_file = args.test
features, labels, pivots = Data.load_data(input_file, parquet=args.parquet)
test_features, test_labels, test_pivots = Data.load_data(test_file, parquet=args.parquet)
# Build graph
model = Model(config, features=features, labels=labels, args=args)
saver = tf.train.Saver()
with tf.Session(config=tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)) as sess:
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
train_handle = sess.run(model.train_iterator.string_handle())
test_handle = sess.run(model.test_iterator.string_handle())
if args.restore_last and ckpt.model_checkpoint_path:
# Continue training saved model
saver.restore(sess, ckpt.model_checkpoint_path)
print('{} restored.'.format(ckpt.model_checkpoint_path))
start_epoch = args.restart_epoch
else:
if args.restore_path:
new_saver = tf.train.import_meta_graph('{}.meta'.format(args.restore_path))
new_saver.restore(sess, args.restore_path)
print('{} restored.'.format(args.restore_path))
start_epoch = args.restart_epoch
else:
start_epoch = 0
sess.run(model.test_iterator.initializer, feed_dict={
model.test_features_placeholder:test_features,
model.test_pivots_placeholder:test_pivots,
model.pivots_placeholder:test_pivots,
model.test_labels_placeholder:test_labels})
for epoch in range(start_epoch, config.num_epochs):
sess.run(model.train_iterator.initializer, feed_dict={model.features_placeholder:features,
model.labels_placeholder:labels, model.pivots_placeholder:pivots})
v_auc_best = Utils.run_diagnostics(model, config, directories, sess, saver, train_handle,
test_handle, start_time, v_auc_best, epoch, global_step, args.name, v_cvm)
if epoch > 0:
save_path = saver.save(sess, os.path.join(directories.checkpoints,
'MI_reg_{}_epoch{}_step{}.ckpt'.format(args.name, epoch, global_step)), global_step=epoch)
print('Starting epoch {}, Weights saved to file: {}'.format(epoch, save_path))
while True:
try:
# Update weights
global_step, *ops = sess.run([model.global_step, model.opt_op, model.MINE_labels_train_op, model.auc_op, model.update_accuracy],
feed_dict={model.training_phase: True, model.handle: train_handle})
if args.mutual_information_penalty:
for _ in range(args.MI_iterations):
sess.run(model.MINE_train_op, feed_dict={model.training_phase: True, model.handle: test_handle})
if global_step % 1000 == 0:
# Periodically show diagnostics
v_MI_kraskov, v_MI_MINE, v_pred, v_labels, v_pivots, v_conf = sess.run([model.MI_logits_theta_kraskov,
model.MI_logits_theta, model.pred, model.labels, model.pivots[:,0], model.softmax],
feed_dict={model.training_phase: False, model.handle: test_handle})
v_cvm = Utils.cvm_z(v_pivots, v_pred, v_labels, confidence=v_conf, selection_fraction=0.05)
v_auc_best = Utils.run_diagnostics(model, config_train, directories, sess, saver, train_handle,
test_handle, start_time, v_auc_best, epoch, global_step, args.name, v_cvm)
if global_step % 1e5 == 0:
save_path = saver.save(sess, os.path.join(directories.checkpoints,
'MI_reg_{}_epoch{}_step{}.ckpt'.format(args.name, epoch, global_step)), global_step=epoch)
print('Weights saved to file: {}'.format(save_path))
except tf.errors.OutOfRangeError:
print('End of epoch!')
break
except KeyboardInterrupt:
save_path = saver.save(sess, os.path.join(directories.checkpoints,
'MI_reg_{}_last.ckpt'.format(args.name)), global_step=epoch)
print('Interrupted, model saved to: ', save_path)
sys.exit()
save_path = saver.save(sess, os.path.join(directories.checkpoints,
'MI_reg_{}_end.ckpt'.format(args.name)),
global_step=epoch)
print("Training Complete. Model saved to file: {} Time elapsed: {:.3f} s".format(save_path, time.time()-start_time))
def main(**kwargs):
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
# Dataset-related options
dataset = parser.add_argument_group("Dataset-related options")
dataset.add_argument("-i", "--input", default=None, help="Path to training file", type=str)
dataset.add_argument("-test", "--test", default=None, help="Path to test file", type=str)
dataset.add_argument("-pq", "--parquet", help="Use if dataset is in parquet format", action="store_true")
dataset.add_argument("-n", "--name", default="MI_reg", help="Checkpoint/Tensorboard label")
# Optimization-related options
optim = parser.add_argument_group("Optimization-related options")
optim.add_argument("-opt", "--optimizer", default="adam", help="Selected optimizer", type=str)
optim.add_argument("-MI_iters", "--MI_iterations", default=16, help="""Number of gradient steps of the 'discriminator'
per encoder gradient step.""", type=int)
# Penalty-related options
penalty = parser.add_argument_group("Penalty-related options. Note only one type of penalty should be active.")
penalty.add_argument("-MI", "--mutual_information_penalty", help="Penalize mutual information between Z and logits", action="store_true")
penalty.add_argument("-lambda", "--MI_lambda", default=0.0, help="Lagrange multiplier in MI-augmented objective.", type=float)
penalty.add_argument("-JSD", "--JSD", help="Use Jensen-Shannon approximation of mutual information", action="store_true")
penalty.add_argument("-heuristic", "--heuristic", help="Use heuristic cost formulation", action="store_true")
penalty.add_argument("-combined", "--combined", help="Use combined cost formulation", action="store_true")
penalty.add_argument("-kl", "--kl_update", help="Use approximate D_KL minimization.", action="store_true")
# Regularization-based options
reg = parser.add_argument_group("Regularization-related options to stabilize training.")
reg.add_argument("-jsd_reg", "--jsd_regularizer", help="Toggle gradient-based regularization", action="store_true")
reg.add_argument("-sn", "--spectral_norm", help="Apply spectral norm to discriminator", action="store_true")
# Miscellaneous options
misc = parser.add_argument_group("Miscellaneous options.")
misc.add_argument("-rl", "--restore_last", help="Restore last saved model", action="store_true")
misc.add_argument("-r", "--restore_path", help="Path to model to be restored", type=str)
misc.add_argument("-re", "--restart_epoch", default=0, help="Epoch to restart from", type=int)
misc.add_argument("-bkg_only", "--bkg_only", help="Apply penalty only to background class.", action="store_true")
args = parser.parse_args()
# Launch training
train(config_train, args)
if __name__ == '__main__':
main()