Inference

Inference is the process of giving an input to your network and getting a classification or prediction. When the neural network has been trained and deployed in production, we use it, for example, to classify images or to decide how to drive in a road, and this process is called inference.

The first step is to load the model:

model = load_model(os.path.join(dir_save, model_name))

Then you simply call predict(), which is the method for inference in Keras. Let's try with the first test sample of MNIST:

x_pred = model.predict(x_test[0:1, :, :, :])print("Expected:", np.argmax(y_test))print("First prediction probabilities:", x_pred)print("First prediction:", np.argmax(x_pred))

This is the result for my MNIST network:

Expected: 7
First prediction probabilities: [[6.3424804e-14 6.1755254e-06 2.5011676e-08 2.2640785e-07 9.0170204e-08 7.4626680e-11 5.6195684e-13 9.9999273e-01 1.9735349e-09 7.3219508e-07]]
First prediction: 7

...