Using LIME Image Explainer

This example illustrates how to use the LIME image explainer to explain a prediction.

This example shows how to use the image LIME implementation – fatf.transparency.predictions.surrogate_image_explainers.ImageBlimeyLime – to explain a prediction of a black-box probabilistic model.

• 
• 

Out:

{'bottom-left': 0.059825904043669643,
 'bottom-right': -0.27639956335921306,
 'top-left': 0.4983489307756843,
 'top-right': 0.04997774025460273}

# Author: Kacper Sokol <k.sokol@bristol.ac.uk>
# License: new BSD

from pprint import pprint

import matplotlib.pyplot as plt
import numpy as np

import fatf

import fatf.utils.models as fatf_models

import fatf.transparency.predictions.surrogate_image_explainers as fatf_exp

import fatf.vis.lime as fatf_vis_lime

print(__doc__)

# Fix random seed
fatf.setup_random_seed(42)

# Create a simple data set
r, g, b, k = [255, 0, 0], [0, 255, 0], [0, 0, 255], [0, 0, 0]
X = np.array(
    [
        [[r, g], [b, k]],
        [[r, b], [g, k]],
        [[r, k], [b, g]],
        [[k, g], [b, r]],
        [[k, b], [g, r]],
        [[g, k], [b, r]]
    ],
    dtype=np.uint8
)
y = np.array([0, 0, 0, 1, 1, 1])

feature_names = {
  'Segment #1': 'top-left',
  'Segment #2': 'top-right',
  'Segment #3': 'bottom-left',
  'Segment #4': 'bottom-right'
}
class_names = {0: 'top-left-red', 1: 'bottom-right-red'}

# Create a model that operates of 3-D images
class KNNimg(fatf_models.KNN):
    def fit(self, X, y):
        X_r_flat = np.array([i[:, :, 0].flatten() for i in X], dtype=np.uint8)
        super().fit(X_r_flat, y)

    def predict(self, X):
        X_r_flat = np.array([i[:, :, 0].flatten() for i in X], dtype=np.uint8)
        return super().predict(X_r_flat)

    def predict_proba(self, X):
        X_r_flat = np.array([i[:, :, 0].flatten() for i in X], dtype=np.uint8)
        return super().predict_proba(X_r_flat)

# Train a model
clf = KNNimg(k=1)
clf.fit(X, y)

# Choose an instance to be explained
image = X[0]
image_class = 0

# Display the instance
plt.axis('off')
plt.imshow(image)

# Create a LIME image explainer
lime = fatf_exp.ImageBlimeyLime(
    image,
    clf,
    ratio=1,
    kernel_size=1.05,
    max_dist=10,
    colour='random')

# Explain an instance
lime_explanation_ = lime.explain_instance(explained_class=image_class)

# Remap the explanation
lime_explanation = {feature_names[i]: j for i, j in lime_explanation_.items()}

# Display the textual explanation
pprint(lime_explanation)

# Plot the explanation
fatf_vis_lime.plot_lime(lime_explanation)