model_train.py

# -*- coding: utf-8 -*-
# @Time    : 2020-11-05 17:41
# @Author  : WenYi
# @Contact : 1244058349@qq.com
# @Description :  script description


import tensorflow as tf
import os
import time

tf.config.set_soft_device_placement(True)
tf.debugging.set_log_device_placement(True)
from tensorflow.keras import optimizers
from tensorflow.keras.callbacks import *
from esmm import CTCVRNet

os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = "true"


def train_model(cate_feature_dict, user_cate_feature_dict, item_cate_feature_dict, train_data, val_data):
	"""
	model train and save as tf serving model
	:param cate_feature_dict: dict, categorical feature for data
	:param user_cate_feature_dict: dict, user categorical feature
	:param item_cate_feature_dict: dict, item categorical feature
	:param train_data: DataFrame, training data
	:param val_data: DataFrame, valdation data
	:return: None
	"""
	ctcvr = CTCVRNet(cate_feature_dict)
	ctcvr_model = ctcvr.build(user_cate_feature_dict, item_cate_feature_dict)
	opt = optimizers.Adam(lr=0.003, decay=0.0001)
	ctcvr_model.compile(optimizer=opt, loss=["binary_crossentropy", "binary_crossentropy"], loss_weights=[1.0, 1.0],
	                    metrics=[tf.keras.metrics.AUC()])
	
	# keras model save path
	filepath = "esmm_best.h5"
	
	# call back function
	checkpoint = ModelCheckpoint(
		filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='min')
	reduce_lr = ReduceLROnPlateau(
		monitor='val_loss', factor=0.8, patience=2, min_lr=0.0001, verbose=1)
	earlystopping = EarlyStopping(
		monitor='val_loss', min_delta=0.0001, patience=8, verbose=1, mode='auto')
	callbacks = [checkpoint, reduce_lr, earlystopping]
	
	# load data
	ctr_user_numerical_feature_train, ctr_user_cate_feature_train, ctr_item_numerical_feature_train, \
	ctr_item_cate_feature_train, cvr_user_numerical_feature_train, cvr_user_cate_feature_train, \
	cvr_item_numerical_feature_train, cvr_item_cate_feature_train, ctr_target_train, cvr_target_train = train_data
	
	ctr_user_numerical_feature_val, ctr_user_cate_feature_val, ctr_item_numerical_feature_val, \
	ctr_item_cate_feature_val, cvr_user_numerical_feature_val, cvr_user_cate_feature_val, \
	cvr_item_numerical_feature_val, cvr_item_cate_feature_val, ctr_target_val, cvr_target_val = val_data
	
	# model train
	ctcvr_model.fit([ctr_user_numerical_feature_train, ctr_user_cate_feature_train, ctr_item_numerical_feature_train,
	                 ctr_item_cate_feature_train, cvr_user_numerical_feature_train, cvr_user_cate_feature_train,
	                 cvr_item_numerical_feature_train,
	                 cvr_item_cate_feature_train], [ctr_target_train, cvr_target_train], batch_size=256, epochs=50,
	                validation_data=(
		                [ctr_user_numerical_feature_val, ctr_user_cate_feature_val, ctr_item_numerical_feature_val,
		                 ctr_item_cate_feature_val, cvr_user_numerical_feature_val, cvr_user_cate_feature_val,
		                 cvr_item_numerical_feature_val,
		                 cvr_item_cate_feature_val], [ctr_target_val, cvr_target_val]), callbacks=callbacks,
	                verbose=0,
	                shuffle=True)
	
	# load model and save as tf_serving model
	saved_model_path = './esmm/{}'.format(int(time.time()))
	ctcvr_model = tf.keras.models.load_model('esmm_best.h5')
	tf.saved_model.save(ctcvr_model, saved_model_path)