AssertionError: assert not np.isnan(cost). How to solve this NAN error. Thanks in advance.

Question

AssertionError: assert not np.isnan(cost). How to solve this NAN error. Thanks in advance.

ViswanathaReddyGajjala opened this issue 6 years ago · 8 comments

ViswanathaReddyGajjala commented 6 years ago

Answer 1 · 2019-04-19T14:00:58.000Z

Which notebook is this?

Answer 2 · 2019-04-20T07:38:07.000Z

https://github.com/huseinzol05/NLP-Models-Tensorflow/blob/master/text-classification/65.gpt-2.ipynb

The notebook is for binary classification. I have modified the code for multi-class classification. But, the accuracy is around 55.5 . By using BERT, I am getting around 75. Is it ok to have such a large margin between those two models? Thanks in advance.

Answer 3 · 2019-04-20T15:25:24.000Z

I am also wondering why GPT-2 is not really powerful, maybe my code is not really a good in structure, I might need to check the GPT-2 inside tensorboard

Answer 4 · 2019-04-20T16:47:41.000Z

Have you trained the model in an unsupervised manner?

Answer 5 · 2019-04-21T13:27:01.000Z

can you elaborate more about unsupervised manner?

Answer 6 · 2019-04-21T13:28:11.000Z

Are you using the pre-trained model released by OpenAI?

Answer 7 · 2019-04-21T15:49:55.000Z

nope, I just took the model code, and train it from scratch. Maybe you can try to do and let me know the result.

Answer 8 · 2019-04-23T05:10:38.000Z

Hi, GPT-2 from OpenAI is not full data they release, as they say, maybe that is the answer