Tensorflow 深度学习问题总结

1、报错信息：tensorflow.python.framework.errors_impl.InvalidArgumentError:  assertion failed: [predictions must be >= 0] [Condition x >= y did not hold element-wise:] [x (sub_2:0) = ] [[-nan][-nan][-nan]...] [y (Cast_2/x:0) = ] [0]
         [[{{node assert_greater_equal/Assert/AssertGuard/else/_1/assert_greater_equal/Assert/AssertGuard/Assert}}]] [Op:__inference_train_function_1223]

Function call stack:
train_function

问题原因：tf.sqrt（）开根号导致在0处不可导NAN，可能出现梯度爆炸问题

解决方法：加个很小的数，或者将学习率调小

2、警告信息：tensorflow:Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least `steps_per_epoch * epochs` batches (in this case, 10000 batches). You may need to use the repeat() function when building your dataset.

问题原因：batch_size和steps_per_epoch、validation_steps同时出现出现冲突

解决方法：保留batch_size或者steps_per_epoch、validation_step

3、The `batch_size` argument must not be specified for the given input type. Received input: <BatchData....

问题原因：可能数据集制作的时候就使用了.batch(64)做好了，就不用在fit中加入batch_size了？【纯属猜测，因为在自己环境跑没出现问题，而是在别人的平台上跑出现的问题】

解决方法：去掉batch_size参数