pca.py

compressed_image_data_train = pca.transform(image_data_train)
uncompressed_image_data_train = pca.inverse_transform(compressed_image_data_train)

fig=plt.figure(figsize=(12, 20))
columns = 4
rows = 8
for i in range(1, 33, 4):
    component_id = int(i/4)
    eigen_image = image_from_component_values(pca.components_[component_id])
    inverted_egein_image = PIL.ImageOps.invert(eigen_image)
    
    most_similar_id = (np.argmax(compressed_image_data_train[:, component_id]))
    most_dissimilar_id = (np.argmin(compressed_image_data_train[:, component_id]))

    
    fig.add_subplot(rows, columns, i)
    plt.imshow(eigen_image)
    
    fig.add_subplot(rows, columns, i+1)
    similar_image = plt.imshow(
        raw_image_data[most_similar_id, :].reshape(IMAGE_SIZE[1], IMAGE_SIZE[0], 3))
    
    fig.add_subplot(rows, columns, i+2)
    similar_image = plt.imshow(
        raw_image_data[most_dissimilar_id, :].reshape(IMAGE_SIZE[1], IMAGE_SIZE[0], 3))

    
    fig.add_subplot(rows, columns, i+3)
    plt.imshow(inverted_egein_image)
    
plt.show()