plot_feature_importance got stuck

[mattivi]

I am running the plot_feature_importance in a jupyter notebook (python 3.6, skater-1.1.2) but it got stuck without providing any error. Code is the following

interpreter = Interpretation(training_data=X_train_XAI, feature_names=data.drop(columns=col_emb).columns)
im_model = InMemoryModel(predict_AE_XAI_API, examples=X_real_XAI)
plots = interpreter.feature_importance.plot_feature_importance(im_model, ascending=True, n_samples=20000, use_scaling=True, n_jobs=10)

the interpreter.feature_importance.plot_feature_importance is running endlessly, without consuming CPU resources. How to debug it and find why it get stuck? Or any clue why this behaviour? Many thanks!

[marcayda]

Hey Hi, Do you have any solutions for this? I have the same problem :S

[mattivi]

No. It works only if I reduce the number of samples to 15000, so in my case X_real_XAI has a dimension of (15000,7). However, that's not the purpose of explainability as I need to evaluate for all rows in the original dataset, that in my case are much higher.

[mattivi]

Any more hint how to solve this problem?

[marcayda]

I installed Skater with Conda and I don't have this problem anymore, but I can't use de feature n_samples, I see the next message in console TypeError: plot_feature_importance() got an unexpected keyword argument 'n_samples', I don't use de Jpyter Editor, maybe is for this reason the error message.

In this link is the description of the install with Conda

https://oracle.github.io/Skater/install.html

[bonogom]

I got the same problem and I sovled it with specifying n_job (n_job=1). I guess this issue is related to multiprocessing.