CS3244-Project/train_cnn.py at master · Lingy12/CS3244-Project · GitHub

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
# Tools import
import glob
import os
from typing import Tuple
import tensorflow as tf
import numpy as np
from itertools import product
from generate_ct_data import save_np_arr_with_channel

# Layers and Models import
from tensorflow.keras.layers import Input, Conv3D, MaxPool3D, BatchNormalization, Flatten, Dense, Dropout
from tensorflow.keras import Model

# Call back import
from tensorflow.keras.callbacks import CSVLogger, ModelCheckpoint, EarlyStopping

def build_baseline_model(input_shape):
    input_layer = Input(input_shape)
    conv_layer1 = Conv3D(filters=64, kernel_size=(3, 3, 3), activation='relu', padding='same')(input_layer)
    pooling_layer1 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer1)
    pooling_layer1 = BatchNormalization()(pooling_layer1)
    conv_layer2 = Conv3D(filters=64, kernel_size=(3, 3, 3), activation='relu', padding='same')(pooling_layer1)
    pooling_layer2 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer2)
    pooling_layer2 = BatchNormalization()(pooling_layer2)

    pooling_layer9 = BatchNormalization()(pooling_layer2)
    flatten_layer = Flatten()(pooling_layer9)

    dense_layer3 = Dense(units=512, activation='relu')(flatten_layer)
    dense_layer3 = Dropout(0.4)(dense_layer3)

    dense_layer4 = Dense(units=256, activation='relu')(dense_layer3)
    dense_layer4 = Dropout(0.4)(dense_layer3)

    output_layer = Dense(units=1, activation='linear')(dense_layer4)

    model = Model(inputs=input_layer, outputs=output_layer)
    return model

def build_cnn_model(input_shape):
    input_layer = Input(input_shape)
    conv_layer1 = Conv3D(filters=64, kernel_size=(3, 3, 3), activation='relu', padding='same')(input_layer)
    pooling_layer1 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer1)
    pooling_layer1 = BatchNormalization()(pooling_layer1)
    conv_layer2 = Conv3D(filters=64, kernel_size=(3, 3, 3), activation='relu', padding='same')(pooling_layer1)
    pooling_layer2 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer2)
    pooling_layer2 = BatchNormalization()(pooling_layer2)
    conv_layer3 = Conv3D(filters=64, kernel_size=(3, 3, 3), activation='relu', padding='same')(pooling_layer1)
    pooling_layer3 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer3)
    pooling_layer3 = BatchNormalization()(pooling_layer3)
    conv_layer4 = Conv3D(filters=128, kernel_size=(3, 3, 3), activation='relu', padding='same')(pooling_layer3)
    pooling_layer4 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer4)
    pooling_layer4 = BatchNormalization()(pooling_layer4)
    conv_layer5 = Conv3D(filters=256, kernel_size=(3, 3, 3), activation='relu', padding='same')(pooling_layer4)
    pooling_layer5 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer5)
    conv_layer6 = Conv3D(filters=128, kernel_size=(3, 3, 3), activation='relu', padding='same')(pooling_layer5)
    pooling_layer6 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer6)
    conv_layer7 = Conv3D(filters=32, kernel_size=(3, 3, 3), activation='relu', padding='same')(pooling_layer6)
    pooling_layer7 = MaxPool3D(pool_size=(2, 2, 2), padding='same')(conv_layer7)

    pooling_layer9 = BatchNormalization()(pooling_layer7)
    flatten_layer = Flatten()(pooling_layer9)

    dense_layer3 = Dense(units=512, activation='relu')(flatten_layer)
    dense_layer3 = Dropout(0.4)(dense_layer3)

    dense_layer4 = Dense(units=256, activation='relu')(dense_layer3)
    dense_layer4 = Dropout(0.4)(dense_layer3)

    output_layer = Dense(units=1, activation='linear')(dense_layer4)

    model = Model(inputs=input_layer, outputs=output_layer)
    return model

def build_ds_from(input_path, label_path):
    print('Building dataset...')
    label = np.load(label_path)
    print(len(label))
    data = np.load(input_path)
    data = np.expand_dims(data, -1)

    return tf.data.Dataset.from_tensor_slices((data, label))

def generate_training_combination(input_path_list, label_path_list):
    return product(input_path_list, label_path_list)

def batch_dataset(ds, batch_size=1):
    return ds.batch(batch_size)

def split_dataset(dataset: tf.data.Dataset,
                  dataset_size: int,
                  train_ratio: float,
                  validation_ratio: float, test_ratio:float) -> Tuple[tf.data.Dataset, tf.data.Dataset, tf.data.Dataset]:
    assert (train_ratio + validation_ratio) <= 1

    train_count = int(dataset_size * train_ratio)
    validation_count = int(dataset_size * validation_ratio)
    test_count = dataset_size - (train_count + validation_count)

    dataset = dataset.shuffle(dataset_size)

    train_dataset = dataset.take(train_count)
    validation_dataset = dataset.skip(train_count).take(validation_count)
    test_dataset = dataset.skip(validation_count + train_count).take(test_count)

    return batch_dataset(train_dataset), batch_dataset(validation_dataset), batch_dataset(test_dataset)

# Data Access
channel_size = [3, 5, 10, 15, 20, 30, 50, 70, 90]
label_path_list = ['./avg_change.npy']
dataset_size = 174

# tf.debugging.set_log_device_placement(True)
# Training parameter

for (channel_num, label_path) in generate_training_combination(channel_size, label_path_list):
    input_path = f'./ct_interpolated_{channel_num}.npy'
    if not os.path.exists(input_path):
        save_np_arr_with_channel(channel_num)

    ds = build_ds_from(input_path, label_path)
    input_shape = ds.element_spec[0].shape

    print(input_shape)

    input_name = os.path.basename(input_path).split('.')[0]
    label_name = os.path.basename(label_path).split('.')[0]

    model_path = f'./model_checkpoints/{input_name}_{label_name}'
    csv_log_path = f'./train_logs/{input_name}_{label_name}'

    if os.path.exists(model_path):
        continue

    # gpus = tf.config.list_logical_devices('GPU')
    # strategy = tf.distribute.MirroredStrategy(gpus)
    # with strategy.scope():
    # Build and compile model
    model = build_cnn_model((input_shape[0], input_shape[1], input_shape[2], input_shape[3]))
    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.01),
            loss=tf.keras.losses.MeanSquaredError(),
            metrics=[tf.keras.metrics.MeanSquaredError(),
                    tf.keras.metrics.MeanAbsoluteError()])

    # Setup callbacks
    csv_logger_callback = CSVLogger(f'./train_logs/{input_name}_{label_name}')
    checkpoint_callback = ModelCheckpoint(filepath = f'./model_checkpoints/{input_name}_{label_name}',
                                        monitor='loss',
                                        save_best_only=True)
    earlystop_callback = EarlyStopping(monitor='loss', min_delta=1, patience=200)

    train_ds, val_ds, test_ds = split_dataset(ds, dataset_size, 0.8, 0.1, 0.1)

    model.fit(train_ds, validation_data = val_ds, epochs=1000, callbacks=[csv_logger_callback, checkpoint_callback, earlystop_callback])