Implementing Lime with Sequential three-D Keras model

Question

I want to implement LIME explainer to describe a three-dimensional Keras Sequential model (n_samples, n_timesteps, n_features).

I cannot understand whether I have to use Lime "RecurrentTabularExplainer" or "LimeTabularExplainer".

Also I cannot understand what the "classifier_fn" in "explain_instance" has to return and in what size.

For the moment I'm proceeding like this:

explainer = lime_tabular.RecurrentTabularExplainer(
      X_train
    , training_labels      = y_train
    , feature_names        = features
    , categorical_features = [3,5]
    , class_names          = ['Ko', 'Ok']
    , mode                 = 'regression'
)

i = 40
exp = explainer.explain_instance(
      X_test[ i ]
    , classifier_f
    , labels = y_test[ i ]
)

with

X_train.shape (4124,60,19)
y_train.shape (4124,60)

X_test.shape (938,60,19)
y_test.shape (938,60)

Model

 Layer (type)                Output Shape              Param #   
=================================================================
 masking (Masking)           (None, 60, 19)            0         
                                                                 
 gru (GRU)                   (None, 60, 100)           25800     
                                                                 
 batch_normalization (BatchN  (None, 60, 100)          200       
 ormalization)                                                   
                                                                 
 time_distributed (TimeDistr  (None, 60, 1)            101       
 ibuted)                                                         
=================================================================

Thank you a lot in advance.

Implementing Lime with Sequential three-D Keras model

0 Answers0