"Domain error in arguments" exception after lime update

Hey,

I updated lime package from lime==0.1.1.34 to lime==0.1.1.35 version. After that LimeTabularExplainer raised ValueError: Domain error in arguments.

explainer = LimeTabularExplainer(
    training_data=X_train.toarray(),
    mode='classification',
    feature_names=feature_names,
    class_names=[0, 1],
    feature_selection='highest_weights',
    random_state=42
)

X_train object is a sparse matrix and feature_names is a list of strings.

• Python 3.6.8 x64
• numpy 1.16.4
• scipy 1.3.0

Exception details:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-22-68c347456caa> in <module>
      8     ],
      9     feature_selection='highest_weights',
---> 10     random_state=42
     11 )

/project/env/lib/python3.6/site-packages/lime/lime_tabular.py in __init__(self, training_data, mode, training_labels, feature_names, categorical_features, categorical_names, kernel_width, kernel, verbose, class_names, feature_selection, discretize_continuous, discretizer, sample_around_instance, random_state, training_data_stats)
    214                 self.discretizer = QuartileDiscretizer(
    215                         training_data, self.categorical_features,
--> 216                         self.feature_names, labels=training_labels)
    217             elif discretizer == 'decile':
    218                 self.discretizer = DecileDiscretizer(

/project/env/lib/python3.6/site-packages/lime/discretize.py in __init__(self, data, categorical_features, feature_names, labels, random_state)
    185         BaseDiscretizer.__init__(self, data, categorical_features,
    186                                  feature_names, labels=labels,
--> 187                                  random_state=random_state)
    188 
    189     def bins(self, data, labels):

/project/env/lib/python3.6/site-packages/lime/discretize.py in __init__(self, data, categorical_features, feature_names, labels, random_state, data_stats)
     97             self.maxs[feature] = qts.tolist() + [boundaries[1]]
     98             [self.get_undiscretize_value(feature, i)
---> 99              for i in range(n_bins + 1)]
    100 
    101     @abstractmethod

/project/env/lib/python3.6/site-packages/lime/discretize.py in <listcomp>(.0)
     97             self.maxs[feature] = qts.tolist() + [boundaries[1]]
     98             [self.get_undiscretize_value(feature, i)
---> 99              for i in range(n_bins + 1)]
    100 
    101     @abstractmethod

/project/env/lib/python3.6/site-packages/lime/discretize.py in get_undiscretize_value(self, feature, val)
    136                 scipy.stats.truncnorm.rvs(
    137                     minz, maxz, loc=means[val], scale=stds[val],
--> 138                     random_state=self.random_state, size=self.precompute_size))
    139         idx = self.undiscretize_idxs[feature][val]
    140         ret = self.undiscretize_precomputed[feature][val][idx]

/project/env/lib/python3.6/site-packages/scipy/stats/_distn_infrastructure.py in rvs(self, *args, **kwds)
    960         cond = logical_and(self._argcheck(*args), (scale >= 0))
    961         if not np.all(cond):
--> 962             raise ValueError("Domain error in arguments.")
    963 
    964         if np.all(scale == 0):

ValueError: Domain error in arguments.

[knchanu]

@ronykroy yes, setting the parameter, discretize_continuous=False solved this issue for me.

[ronykroy]

hi there… Could you try with setting the discretize_continuous=True parameter to False… my use case: newsGroups20 dataset h20 w2v vectorizer 200 cols values in the 200 cols after vectorization is almost continuous, going by the name of the parameter discretize_continuous … it doesnt make sense to set it to true here…

thats my guess… a better explanation is welcome though