search

tensorflow / lucid

A collection of infrastructure and tools for research in neural network interpretability.
Apache License 2.0
4.65k stars 655 forks source link

AssertionError in calling render_vis #229

Closed Uiuran closed 4 years ago

Uiuran commented 4 years ago

Tensorflow 1.14.0 Numpy 1.17.4

the following stack just pop-up: https://pastebin.com/mL4TUMxz

since your code calls on TF to get input Tensor, eval it and use it as input for show, and you use numpy to see the rank and stack and is nan and so on ... the bug may be somewhere there.

Also, i noticed you dont put kwargs in render_vis, in such a fashion that i cant pass w keyword arg and make the visualization bigger.

Still wondering why not to use matplotlib with jupyter ...

ps: i finally decided to switch to Lucid to do deep dream and other Artistic things, because it seems to be much more evolved than the initial deep dream code ...

Uiuran commented 4 years ago

will try tf 1.15., however the numpy 1.16 version of the tutorial may crash another libs to downgrade.

Is there a stable version ? and what is 71% of the code coverage means ? thanks in advance ...

Uiuran commented 4 years ago

I cant get ride of the assertion error even with 0.3.8. Even with tf 1.15 and numpy 1.16 Tutorial Notebook wont work.

colah commented 4 years ago

It sounds like you are running into numerical stability issues while visualizing your model. This could be caused by many issues -- for example, the objective you are visualizing could be exploding, your model may have unstable gradients, and much more.

Please keep in mind that lucid is research code.

Uiuran commented 4 years ago

I will close this issue since seems to be the same mentioned earlier, so we keep the discussion in the other issue.