plot_lime_top_explanations.py

def plot_lime_top_explanations(
    model: Model, 
    image: np.array, 
    class_names_mapping: Dict[int, str],
    top_preds_count: int = 3,
    fig_name: Optional[str] = None
) -> None:
    
    image_columns = 3
    image_rows = math.ceil(top_preds_count / image_columns)
    
    explanation = explainer.explain_instance(
        image, 
        classifier_fn = model.predict, 
        top_labels=100, 
        hide_color=0, 
        num_samples=1000
    )
    
    preds = model.predict(np.expand_dims(image, axis=0))
    top_preds_indexes = np.flip(np.argsort(preds))[0,:top_preds_count]
    top_preds_values = preds.take(top_preds_indexes)
    top_preds_names = np.vectorize(lambda x: class_names[x])(top_preds_indexes)
    
    plt.style.use('dark_background')
    fig, axes = plt.subplots(image_rows, image_columns, figsize=(image_columns * 5, image_rows * 5))
    [ax.set_axis_off() for ax in axes.flat]
    
    for i, (index, value, name, ax) in \
        enumerate(zip(top_preds_indexes, top_preds_values, top_preds_names, axes.flat)):
        
        temp, mask = explanation.get_image_and_mask(
            explanation.top_labels[i], 
            positive_only=False, 
            num_features=5, 
            hide_rest=False
        )
        
        subplot_title = "{}. class: {} pred: {:.3f}".format(i + 1, name, value)
        ax.imshow(mark_boundaries(temp / 255, mask))
        ax.set_title(subplot_title, pad=20)
    
    
    if fig_name:
        plt.savefig(fig_name)
    plt.show()