NN Tuning

def build_model(learning_rate, hidden_nodes = 64, dropout_rate = 0.2, activation_function = tf.nn.relu):
  model = keras.Sequential([
     layers.Dense(hidden_nodes, activation=activation_function, input_shape=[len(train_dataset.keys())]),
     layers.Dropout(dropout_rate),
     layers.Dense(hidden_nodes, activation=activation_function),
     layers.Dense(1)
  ])

  optimizer = tf.keras.optimizers.RMSprop(learning_rate)

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



from keras.wrappers.scikit_learn import KerasRegressor
from sklearn.model_selection import RandomizedSearchCV

EPOCHS = 1000

learning_rates = [0.001, 0.01, 0.1, 0.2, 0.3]
hidden_nodes = [16, 32, 40, 64, 80]
dropout_rates = [0.1, 0.2, 0.3, 0.4]
activation_functions = [tf.nn.relu, tf.nn.leaky_relu]

hyperparameters = dict(learning_rate = learning_rates, hidden_nodes = hidden_nodes, dropout_rate = dropout_rates, activation_function = activation_functions)
hyperparameters

model = KerasRegressor(build_fn = build_model, epochs=EPOCHS)
random_search = RandomizedSearchCV(estimator = model, param_distributions = hyperparameters)
random_search.fit(train_dataset, train_labels)