Create Neural Network model

The model creation procedure always looks the same:

  1. Initialize model with the expected shape of the input data

model = Model(input_shape=(1, 2))

  1. Add some layers to the model

model.add(Dense(3))
model.add(Dense(1))

  1. Compile model.

Warning

Without this step, the model will not be able to train and predict.

model.compile(
    optimizer='gradient_descent',
    loss='mean_squared_error',
    metrics=['mean_absolute_error']
)

Now, the model is ready for training and prediction. For training, two options for printing results are possible (see the verbosity parameter of the Model.fit() function). In most cases, your code will look something like this:

model = Model(input_shape=(1, 2))

model.add(Dense(3, activation='sigmoid'))
model.add(Dense(1, activation='sigmoid'))

model.compile(
    optimizer='gradient_descent',
    loss='mean_squared_error',
    metrics=['mean_absolute_error']
)

x_train = np.array([[[0, 0]], [[0, 1]], [[1, 0]], [[1, 1]]])
y_train = np.array([[[0]], [[1]], [[1]], [[0]]])

model.fit(
    x_train, y_train,
    validation_data=[x_train, y_train],
    batch_size=4,
    epochs=10000,
    verbosity=1  # progress bar
)

For prediction, two options are possible:

  • Pass only the input data (x_test) to the function. In this case, the function will return only the predicted data.

model.predict(x_test)

  • Pass both x_test and y_test data. In this case, the function will return the predicted data and print the estimated metrics for that data

model.predict(x_test, y_test)

More examples can be found in github repository: