-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
144 lines (118 loc) · 4.61 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
import os
import tensorflow as tf
from tensorflow.keras.callbacks import ModelCheckpoint
import tensorflow_datasets as tfds
import argparse
import yaml
import models
def normalize_img(image, label):
"""
Normalizes images: `uint8` -> `float32`.
"""
return tf.cast(image, tf.float32) / 255., label
def expand_img_dim(image, label):
"""
Make sure images have shape (28, 28, 1)
"""
return tf.expand_dims(image, -1), label
def main(args):
with open(args.config) as file:
our_config = yaml.safe_load(file)
save_dir = our_config['save_dir']
os.makedirs(save_dir, exist_ok=True)
model_name = our_config['model']['name']
model_file_path = os.path.join(save_dir, 'model_best.h5')
num_epochs = our_config['model']['epochs']
# Prepare dataset
(ds_train, ds_test), ds_info = tfds.load(
'mnist',
split=['train', 'test'],
shuffle_files=True,
as_supervised=True,
with_info=True,
)
ds_train = ds_train.map(
normalize_img,
num_parallel_calls=tf.data.AUTOTUNE
)
if 'conv2d' in model_name:
# Make sure images have shape (28, 28, 1)
ds_train = ds_train.map(expand_img_dim)
print(ds_train)
ds_train = ds_train.cache()
ds_train = ds_train.shuffle(ds_info.splits['train'].num_examples)
ds_train = ds_train.batch(128)
ds_train = ds_train.prefetch(tf.data.AUTOTUNE)
ds_test = ds_test.map(
normalize_img,
num_parallel_calls=tf.data.AUTOTUNE
)
if 'conv2d' in model_name:
# Make sure images have shape (28, 28, 1)
ds_test = ds_test.map(expand_img_dim)
ds_test = ds_test.batch(128)
ds_test = ds_test.cache()
ds_test = ds_test.prefetch(tf.data.AUTOTUNE)
# quantization parameters
if 'quantized' in model_name:
logit_total_bits = our_config["quantization"]["logit_total_bits"]
logit_int_bits = our_config["quantization"]["logit_int_bits"]
activation_total_bits = our_config["quantization"]["activation_total_bits"]
activation_int_bits = our_config["quantization"]["activation_int_bits"]
print(f"Logit total bits = {logit_total_bits}, int bits = {logit_int_bits}")
print(f"Activation total bits = {activation_total_bits}, int bits = {activation_int_bits}")
model = getattr(models, model_name)(
logit_total_bits,
logit_int_bits,
activation_total_bits,
activation_int_bits
)
else:
model = getattr(models, model_name)()
print(model)
# print model summary
print('#################')
print('# MODEL SUMMARY #')
print('#################')
print(model.summary())
print('#################')
tf.keras.utils.plot_model(model,
to_file=f'{save_dir}/{model_name}.png',
show_shapes=True,
show_dtype=True,
show_layer_names=True,
rankdir="TB",
expand_nested=False)
model.compile(
optimizer=tf.keras.optimizers.Adam(0.001),
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=[tf.keras.metrics.SparseCategoricalAccuracy()],
)
# TODO: lol to set BER probably can do something like
# for layer in model.layers:
# if type(layer) == FQDense:
# ...
# train_ber = 1.0
# print(f"Training with ber = {train_ber}")
# model.get_layer('fq_dense').set_ber(train_ber)
# model.get_layer('fq_dense_1').set_ber(train_ber)
# print(f"Layer {model.get_layer('fq_dense')} has ber = {model.get_layer('fq_dense').get_ber()}")
# print(f"Layer {model.get_layer('fq_dense_1')} has ber = {model.get_layer('fq_dense_1').get_ber()}")
model.fit(
ds_train,
epochs=num_epochs,
validation_data=ds_test,
callbacks=[ModelCheckpoint(model_file_path, monitor='val_loss', verbose=True, save_best_only=True)]
)
# eval_ber = 1.0
# model.get_layer('fq_dense').set_ber(eval_ber)
# model.get_layer('fq_dense_1').set_ber(eval_ber)
# print("Evaluting model with ber = 1.0")
# print(f"Layer {model.get_layer('fq_dense')} has ber = {model.get_layer('fq_dense').get_ber()}")
# print(f"Layer {model.get_layer('fq_dense_1')} has ber = {model.get_layer('fq_dense_1').get_ber()}")
model.evaluate(ds_test)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('-c', '--config', type=str, default="baseline.yml", help="specify yaml config")
args = parser.parse_args()
main(args)