tensorflow 梯度更新对输入：输入转为Variables
optimizer.apply_gradients([gradient，inputs])
optimizer.aplly_gradiets(zip(gradient,inputs))
inputs=inputs+gradient*learning_rate