tfserve_vision10.py

# save model
if not os.path.isdir('model_weights/'):
    os.mkdir('model_weights/')
model2.save_weights(filepath='model_weights/cnn_model2_wt.h5', overwrite=True)

# load model (can be used in the future as needed once trained)
model2 = create_cnn_architecture_model2(input_shape=INPUT_SHAPE_RN)
model2.load_weights('model_weights/cnn_model2_wt.h5')

# predict and evaluate on test dataset
test_images_3ch_scaled = test_images_3ch / 255.
predictions = model2.predict(test_images_3ch_scaled)
prediction_labels = np.argmax(predictions, axis=1)
print(classification_report(test_labels, prediction_labels, 
                            target_names=class_names))
pd.DataFrame(confusion_matrix(test_labels, prediction_labels), 
             index=class_names, columns=class_names)