dandolodavid / acme Goto Github PK

Hello,

Is it possible that the local explanation contains a bug? The following appears for me, although a series was also passed in the example notebook (as it also says in the documentation). Regarding the code, I stick to the demo notebook.

Am I missing something or is this a bug?

Reference: https://github.com/dandolodavid/ACME/blob/master/notebook/ACME%20classification.ipynb

Error:
....
in predictACME(model, series, features, percentile_functions, meta_table, task, score_function, class_to_analyze)
348 Z_list.append(baseline)
349 meta_list.append({'feature':feature_table.index.unique()[0],
350 'quantile':get_exact_quantile(percentile_functions[feature],baseline[feature])[0],
...
5900 ):
5901 return self[name]
-> 5902 return object.getattribute(self, name)

AttributeError: 'Series' object has no attribute 'iterrows'

Cheers

n_features only works with local summary plot. We must extend it to global summary plot also

Could be useful to have a parameter that specify if the feature order in the plot should be the global importance or the local importance.