train.py

#!/usr/bin/env python
# -*- encoding: utf-8 -*-
"""
@File    :   train.py
@Contact :   huanghoward@foxmail.com
@Modify Time :    2021/11/10 15:38  
------------      
"""
import time

import numpy as np
from keras.layers import Input
from keras.models import Model
from matplotlib import pyplot as plt
from sklearn.model_selection import train_test_split

from model import *

# 记录脚本运行时间
time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime())

# 加载数据
X = np.load(r"tsp_data/tsp_5/tsp5_train_x.npy")
Y = np.load(r"tsp_data/tsp_5/tsp5_train_y.npy")
x_test = np.load(r"tsp_data/tsp_5/tsp5_test_x.npy")
y_test = np.load(r"tsp_data/tsp_5/tsp5_test_y.npy")
x_train, x_valid, y_train, y_valid = train_test_split(X, Y, test_size=0.2)
print(x_train.shape, y_train.shape)

# 构建模型
main_input = Input(shape=(5, 2), name='main_input')
enc_output, state_h, state_c = Encoder(hidden_dimensions=256)(main_input)
outputs = Decoder(hidden_dimensions=256)(main_input, enc_output, [state_h, state_c])
model = Model(main_input, outputs)
print(model.summary())
# 指定训练配置
model.compile(optimizer='SGD',
              loss='categorical_crossentropy',
              metrics=['accuracy'])
# 训练模型
model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
    filepath='./data/checkpoint-{}.h5'.format(time),
    save_weights_only=True,
    monitor='val_accuracy',
    mode='max',
    save_best_only=True)
early_stop_callback = tf.keras.callbacks.EarlyStopping(
    monitor="val_accuracy",
    min_delta=0,
    patience=12,
    verbose=0,
    mode="auto",
    baseline=None,
    restore_best_weights=False,
)
# 训练
model.load_weights("data/checkpoint-2021-11-18-00-07-32.h5")
history = model.fit(x_train,
                    y_train,
                    epochs=1000,
                    validation_data=(x_valid, y_valid),
                    batch_size=500,
                    callbacks=[model_checkpoint_callback, early_stop_callback])
# 绘制训练 & 验证的准确率值
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Model accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Test'], loc='upper left')
plt.show()

# 绘制训练 & 验证的损失值
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('Model loss')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['Train', 'Test'], loc='upper left')
plt.show()