本文介绍了一种常见的TensorFlow训练错误——训练损失变为NaN的原因及解决方法。作者通过排查发现，错误根源在于输入了错误的标签数据，导致网络无论怎样调整参数都无法得到正确的预测结果。
Days ago, I met the error, training loss goes to be “nan” in tensorflow. Payed some effort, I found the cause of this error. In my case, It was I fed the wrong label to the network caused that error. In the wrong labeled scenario, no matter the direction the network going in training, there are wrongs and maybe more wrongs in prediction.
If you come across the “nan” error in tensorflow, you may need to check the training date you fed in the network first. If there is nothing wrong, go for other investigations.
扫一扫
评论
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
查看更多评论
添加红包
