File size: 1,749 Bytes
2437f34 | 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 | # ====================
# parameter update part
# ====================
from dual_network import DN_INPUT_SHAPE
from tensorflow.keras.callbacks import LearningRateScheduler, LambdaCallback
from tensorflow.keras.models import load_model
from tensorflow.keras import backend as K
from pathlib import Path
import numpy as np
import pickle
NUM_EPOCH = 100
BATCH_SIZE = 128
def load_data():
history_path = sorted(Path('./data').glob('*.history'))[-1]
with history_path.open(mode='rb') as f:
return pickle.load(f)
# Training the dual network
def train_network():
# Loading training data
history = load_data()
s, p, v = zip(*history)
# Reshaping the input data for training
a, b, c = DN_INPUT_SHAPE
s = np.array(s)
s = s.reshape(len(s), c, a, b).transpose(0, 2, 3, 1)
p = np.array(p)
v = np.array(v)
# Loading the best player's model
model = load_model('./model/best.keras')
# Compiling the model
model.compile(loss=['categorical_crossentropy', 'mse'], optimizer='adam')
# Learning rate
def step_decay(epoch):
x = 0.001
if epoch >= 50: x = 0.0005
if epoch >= 80: x = 0.00025
return x
lr_decay = LearningRateScheduler(step_decay)
# Output
print_callback = LambdaCallback(
on_epoch_begin=lambda epoch, logs: print('\rTrain {}/{}'.format(epoch + 1, NUM_EPOCH), end='')
)
# Executing training
model.fit(
s, [p, v], batch_size=BATCH_SIZE , epochs=NUM_EPOCH, verbose=0, callbacks=[lr_decay, print_callback]
)
# Saving the latest player's model
model.save('./model/latest.keras')
# Clearing the model
K.clear_session()
del model
if __name__ == '__main__':
train_network()
|