LIME Explainability with 4D training data on Python

Question

I'm dealing with a Deep learning model requiring as input a 4D array and I would like to use LIME to get some explainability on my results

I'm using lime_tabular.RecurrentTabularExplainer but it requires as training data a 3D numpy array, but my training data are a 4D array.

I tried with this code

from lime.lime_tabular import RecurrentTabularExplainer
explainer = RecurrentTabularExplainer(
    X_train, 
    feature_names=reference_columns, 
    class_names=['No_error', 'Error'], 
    categorical_features=[0, 1]
    #positive_class=0,
    #method='exhaustive'
)

exp = explainer.explain_instance(
    data_row = X_train[0,:,:,:],
    classifier_fn = model_1.predict)

and I obtain the error : "too many values to unpack (expected 3)" since I'm passing a 4D array.

I also tried with this approach. Since my array is shaped as (5240, 57, 37, 1), I reshape it as (5240, 57, 37) when passing it to the RecurrentTabularExplainer

from lime import lime_tabular
data = X_train.reshape(X_train.shape[0], X_train.shape[1], X_train.shape[2])

explainer = lime_tabular.RecurrentTabularExplainer(
    data,
    training_labels = y_train,
    feature_names = reference_columns,
    discretize_continuous = True,
    class_names = ['0','1'],
    discretizer = 'decile')

exp = explainer.explain_instance(
    data_row = X_train[0,:,:,:],
    classifier_fn = model_1.predict)

exp.show_in_notebook()

but then I get this error in the explainer_instance: "ValueError: Input 0 of layer sequential is incompatible with the layer: expected ndim=4, found ndim=3. Full shape received: [None, 57, 37]"

How can i solve this problem?

LIME Explainability with 4D training data on Python

0 Answers0