1、float64 -> float32 耗时下降大概1/5
2、当代码中有多个图时,即建立两个或者多个模型,但是共享数据,会引发错误:
tensorflow.python.framework.errors_impl.InvalidArgumentError: You must feed a value for placeholder tensor 'train_inputs' with dtype int64 and shape [128]
解决参考:https://blog.csdn.net/u012436149/article/details/53894364