-
Notifications
You must be signed in to change notification settings - Fork 27
/
main.py
185 lines (155 loc) · 5.69 KB
/
main.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
import argparse
import os
import numpy as np
import pandas as pd
import tensorflow as tf
import vessl
from sklearn.model_selection import train_test_split
from tensorflow import keras
from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow.python.client import device_lib
from vessl.integration.keras import ExperimentCallback
vessl.init()
def get_available_gpus():
local_device_protos = device_lib.list_local_devices()
return [x.name for x in local_device_protos if x.device_type == "GPU"]
def load_data(data_dir, filename):
data_path = os.path.join(data_dir, filename)
raw_data = pd.read_csv(data_path, dtype=np.float32)
return raw_data
def parse_epoch(file_path):
return int(os.path.splitext(os.path.basename(file_path))[0].split("-")[1])
def preprocess(raw_data):
label = raw_data["label"].to_numpy()
data = raw_data.drop(labels=["label"], axis=1)
data = data / 255.0
data = data.values.reshape(-1, 28, 28)
return label, data
def create_model():
return tf.keras.models.Sequential(
[
tf.keras.layers.Flatten(input_shape=(28, 28)),
tf.keras.layers.Dense(128, activation="relu"),
tf.keras.layers.Dropout(0.2),
tf.keras.layers.Dense(10),
]
)
def save(model, path):
if not os.path.exists(path):
print(f" [*] Make directories : {path}")
os.makedirs(path)
artifact_path = os.path.join(path, "my_model")
model.save(artifact_path)
print(f" [*] Saved model in : {artifact_path}")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Keras MNIST Example")
parser.add_argument(
"--input-path", type=str, default="/input", help="input dataset path"
)
parser.add_argument(
"--output-path", type=str, default="/output", help="output files path"
)
parser.add_argument(
"--checkpoint-path",
type=str,
default="/output/checkpoint",
help="checkpoint file path",
)
parser.add_argument(
"--save-model",
action="store_true",
default=False,
help="save the current model",
)
parser.add_argument(
"--save-model-freq", type=int, default=640, help="save model frequency"
)
parser.add_argument(
"--save-image", action="store_true", default=False, help="save the images"
)
parser.add_argument(
"--vessl-model-repository",
type=str,
default="",
help="Model repository name to save model in VESSL. If not specified, model will not be saved.",
)
args = parser.parse_args()
# hyperparameters
epochs = int(os.environ.get("epochs", 10))
batch_size = int(os.environ.get("batch_size", 128))
optimizer = str(os.environ.get("optimizer", "adam"))
learning_rate = float(os.environ.get("learning_rate", 0.01))
print(f"=> Available GPUs: {get_available_gpus()}")
use_mount_dataset = False
if os.path.exists(os.path.join(args.input_path, "train.csv")) and os.path.exists(
os.path.join(args.input_path, "test.csv")
):
use_mount_dataset = True
if use_mount_dataset:
print("=> Mount dataset found!")
train_df = load_data(args.input_path, "train.csv")
test_df = load_data(args.input_path, "test.csv")
y_train, x_train = preprocess(train_df)
y_test, x_test = preprocess(test_df)
else:
print("=> Mount dataset not found! Use keras dataset instead.")
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
random_seed = 7
x_train, x_val, y_train, y_val = train_test_split(
x_train, y_train, test_size=0.1, random_state=random_seed
)
model = create_model()
print(model.summary())
# Load checkpoint if exists
checkpoint_path = os.path.join(args.checkpoint_path, "checkpoints-{epoch:04d}.ckpt")
checkpoint_dir = os.path.dirname(checkpoint_path)
if os.path.exists(checkpoint_dir) and len(os.listdir(checkpoint_dir)) > 0:
latest = tf.train.latest_checkpoint(checkpoint_dir)
print(f"=> Loading checkpoint '{latest}' ...")
model.load_weights(latest)
start_epoch = parse_epoch(latest)
print(f"start_epoch:{start_epoch}")
else:
start_epoch = 0
if not os.path.exists(args.checkpoint_path):
print(f" [*] Make directories : {args.checkpoint_path}")
os.makedirs(args.checkpoint_path)
checkpoint_callback = ModelCheckpoint(
checkpoint_path,
monitor="val_accuracy",
verbose=1,
save_weights_only=True,
mode="max",
save_freq=args.save_model_freq,
)
# Compile model
if optimizer == "adam":
opt = keras.optimizers.Adam(learning_rate=learning_rate)
elif optimizer == "sgd":
opt = keras.optimizers.SGD(learning_rate=learning_rate)
else:
opt = keras.optimizers.Adadelta(learning_rate=learning_rate)
loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
model.compile(optimizer=opt, loss=loss_fn, metrics=["accuracy"])
model.save_weights(checkpoint_path.format(epoch=0))
model.fit(
x_train,
y_train,
batch_size=batch_size,
validation_data=(x_val, y_val),
epochs=epochs,
callbacks=[
ExperimentCallback(
data_type="image",
validation_data=(x_val, y_val),
num_images=5,
start_epoch=start_epoch,
save_image=args.save_image,
),
checkpoint_callback,
],
)
model.evaluate(x_test, y_test, verbose=2)
if args.save_model:
save(model, args.output_path)