-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
47 lines (35 loc) · 1.43 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
from preprocess import generate_training_sequences, SEQUENCE_LENGTH
from tensorflow import keras
OUTPUT_UNITS = 38 # Tamaño del vocabulario -> Neuronas de la capa de salida
NUM_UNITS = [256] # Neuronas de las capas internas
LOSS = "sparse_categorical_crossentropy"
LEARNING_RATE = 0.001
EPOCHS = 50
BATCH_SIZE = 64 # Muestras que se verán antes del backpropagation
SAVE_MODEL_PATH = "model.h5"
def build_model(input_units, output_units, num_units, loss, learning_rate):
# Creacion de la arquitectura
input = keras.layers.Input(shape=(input_units, output_units))
x = keras.layers.LSTM(num_units[0])(input)
x = keras.layers.Dropout(0.2)(x)
output = keras.layers.Dense(output_units, activation="softmax")(x)
model = keras.Model(input, output)
# Compilación del modelo
model.compile(
loss=loss,
optimizer=keras.optimizers.Adam(learning_rate=learning_rate),
metrics=["accuracy"]
)
model.summary()
return model
def train(input_units=SEQUENCE_LENGTH, output_units=OUTPUT_UNITS, num_units=NUM_UNITS, loss=LOSS, learning_rate=LEARNING_RATE):
# Generar las secuencias de entrenamiento
inputs, targets = generate_training_sequences(SEQUENCE_LENGTH)
# Armar la RNN
model = build_model(input_units, output_units, num_units, loss, learning_rate)
# Entrenar el modelo
model.fit(inputs, targets, epochs=EPOCHS, batch_size=BATCH_SIZE)
# Guardar el modelo
model.save(SAVE_MODEL_PATH)
if __name__ == "__main__":
train(input_units=None)