Shap: No output for force_plot

[Nithanaroy]

import shap

def critical_factors(data_for_prediction):
    shap.initjs()
    background = X_train.iloc[0:100,:].values.astype('float')
    explainer = shap.DeepExplainer(model, background)
    shap_values = explainer.shap_values(data_for_prediction)
    shap.force_plot(explainer.expected_value[0], shap_values[0], data_for_prediction)

critical_factors( X_test.iloc[0:1,:].values.astype('float') )

Just shows a hexagon with JS symbol in it. No plot is visible in the Jupyter Notebook. Do I need to initialize something else?

[Nithanaroy]

Looks like defining a function was the problem. The below works fine. It would be useful to mention this in the README or documentation.

shap.initjs()
data_for_prediction = X_test.iloc[0:1,:].values.astype('float')
background = X_train.iloc[0:100,:].values.astype('float')
explainer = shap.DeepExplainer(model, background)
shap_values = explainer.shap_values(data_for_prediction)
shap.force_plot(explainer.expected_value[0], shap_values[0], data_for_prediction)

However, the plot looses names of the features due to values() of Pandas. How to achieve the same using a DataFrame so that feature names are also visible in the plot?

[slundberg]

The Jupyter notebook only displays the HTML object returned by force_plot because it is the final return value of the last statement in the block. So if you add return to the last statement in the function it should work.

[Nithanaroy]

@slundberg Could you please answer my second comment. Thank you.

[slundberg]

Ah, missed that one. You can just use the feature_names keyword argument to force_plot. It accepts a list of feature names (of the same length as shap_values.shape[1]).