issue with the predict function, classifier problem, categorical dataset

[skavula]

Hi, I found the paper on anchor extremely interesting. The dataset I have only has categorical features with values 0 and 1. I tested it for different models but the code, throws an error in the line, classifier_fn(self.encoder.transform(x)) . As the feature vectors that the dataset has are already discretized, anchor discretizes it further, irrespective of the input throwing an error from the predict function. Could you please help me with the issue. Thanks.

[marcotcr]

Sorry for the delay in responding. Are you using the categorical_features parameter with all of your features when initializing the explainer?

If so, can you please share your code?

[skavula]

No problem. I am using the categorical_features parameter with all the features in the dataset.
The dataset has categorical features only. These features when fed to the model are discretized by anchor, and this doubles the number of feature vectors (the original 51 feature vectors are converted into 102 feature vectors).

clf is a one class SVM model.
exp = explainer.explain_instance(test_data[idx], clf.predict, threshold=0.95)

The error it throws is :
return classifier_fn(self.encoder.transform(x))

ValueError: cannot use sparse input in 'OneClassSVM' trained on dense data

Thank you.

[marcotcr]

I guess I responded to this via email and forgot the thread

[skavula]

Hi Marco, I hope you are doing fine. I had replied to you on github, but I did not receive any response regarding my question. I was hoping if you could help me with the issue. Below is the response to the question you had asked me : I am using the categorical_features parameter with all the features in the dataset. The dataset has categorical features only. These features when fed to the model are discretized by anchor, and this doubles the number of feature vectors (the original 51 feature vectors are converted into 102 feature vectors). [image: screen shot 2018-05-01 at 10 27 38 am] <https://user-images.githubusercontent.com/34525437/39484664-4b85ccfa-4d2b-11e8-92bd-09fbaf24a22e.png> clf is a one class SVM model. exp = explainer.explain_instance(test_data[idx], clf.predict, threshold=0.95) The error it throws is : return classifier_fn(self.encoder.transform(x)) ValueError: cannot use sparse input in 'OneClassSVM' trained on dense data It would be nice if we could take it from here. Thank you, Shaarvani

…

On Fri, May 18, 2018 at 7:18 PM, Marco Tulio Correia Ribeiro < ***@***.***> wrote: I guess I responded to this via email and forgot the thread — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/Ag7Q_ew4eyn8d2xh20hrSqx25CbXgog-ks5tz4DdgaJpZM4TUvd5> .

[marcotcr]

how about this, try encapsulating the SVM function:

def predict_fn(data):
  return clf.predict(data.todense())

And use this in explain_instance