基于tensorflow 的神经网络的损失函数，学习率，正则化，滑动平均等方法(Method of loss function, learning rate, regularization and sliding average of neural network based on tensorflow)