/
model.py
32 lines (21 loc) · 1.38 KB
/
model.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
import tensorflow as tf
def DeepDRP(input_shape=(500, 1910)):
inputs = tf.keras.layers.Input(shape=input_shape, name="inputs")
masking = tf.keras.layers.Masking(input_shape=input_shape, mask_value=0)(inputs)
x = tf.keras.layers.TimeDistributed(tf.keras.layers.Dense(units=1024, activation="tanh"))(masking)
x = tf.keras.layers.TimeDistributed(tf.keras.layers.Dense(units=512, activation="tanh"))(x)
x = tf.keras.layers.TimeDistributed(tf.keras.layers.Dense(units=256, activation="tanh"))(x)
x = tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(units=256, return_sequences=True, activation='tanh'))(x)
x = tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(units=64, return_sequences=True, activation='tanh'))(x)
x = tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(units=2, return_sequences=True, activation='tanh'))(x)
flatten = tf.keras.layers.Flatten()(x)
x = tf.keras.layers.Dense(units=2048, activation="relu")(flatten)
x = tf.keras.layers.Dense(units=1024, activation="relu")(x)
outputs = tf.keras.layers.Dense(units=500, activation="sigmoid")(x)
model = tf.keras.Model(inputs=[inputs], outputs=[outputs])
model.compile(loss=tf.keras.losses.mse,
optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5),
)
return model
model = DeepDRP((500, 1000))
model.summary()