Skip to content

Documentation for KerasDnnRegressor

Wrapper class for a regression-based Keras model.

See Keras API.

Examples:

1
2
3
4
5
6
7
PipelineElement('KerasDnnRegressor',
                hyperparameters={'hidden_layer_sizes': Categorical([[18, 14], [30, 5]]),
                                 'dropout_rate': Categorical([0.01, 0.2])},
                activations='relu',
                epochs=50,
                nn_batch_size=64,
                verbosity=1)
Source code in photonai/modelwrapper/keras_dnn_regressor.py
class KerasDnnRegressor(KerasDnnBaseModel, KerasBaseRegressor):
    """Wrapper class for a regression-based Keras model.

    See [Keras API](https://keras.io/api/).

    Example:
        ``` python
        PipelineElement('KerasDnnRegressor',
                        hyperparameters={'hidden_layer_sizes': Categorical([[18, 14], [30, 5]]),
                                         'dropout_rate': Categorical([0.01, 0.2])},
                        activations='relu',
                        epochs=50,
                        nn_batch_size=64,
                        verbosity=1)
        ```

    """
    def __init__(self,
                 hidden_layer_sizes: int = None,
                 learning_rate: float = 0.01,
                 loss: str = "mean_squared_error",
                 epochs: int = 10,
                 nn_batch_size: int = 64,
                 metrics: list = None,
                 validation_split: float = 0.1,
                 callbacks: list = None,
                 batch_normalization: bool = True,
                 verbosity: int = 0,
                 dropout_rate: Union[float, list] = 0.2,
                 activations: Union[str, list] = 'relu',
                 optimizer: Union[Optimizer, str] = "adam"):
        """
        Initialize the object.

        Parameters:
            hidden_layer_sizes:
                Number of perceptrons per layer.

            learning_rate:
                Step size of the learning adjustment.

            loss:
                Loss function.

            epochs:
                Number of arbitrary cutoffs, generally defined as
                "one pass over the entire dataset", used to separate training into distinct phases,
                which is useful for logging and periodic evaluation.

            nn_batch_size:
                Typically the batch_size. A batch is a set of nn_batch_size samples.
                The samples in a batch are processed independently, in parallel.
                If training, a batch results in only one update to the model.

            metrics:
                List of evaluate metrics.

            callbacks:
                Within Keras, there is the ability to add callbacks specifically designed
                to be run at the end of an epoch. Examples of these
                are learning rate changes and model checkpointing (saving).

            validation_split:
                Split size of validation set.

            batch_normalization:
                Batch normalization applies a transformation that maintains
                the mean output close to 0 and the output standard deviation close to 1.

            verbosity:
                The level of verbosity, 0 is least talkative and
                gives only warn and error, 1 gives adds info and 2 adds debug.

            dropout_rate:
                A Dropout layer applies random dropout and rescales the output.
                In inference mode, the same layer does nothing.
                Float -> added behind each layer
                List -> Same size as hidden_layer_size

            activations:
                Activation function.

            optimizer:
                Optimization algorithm.

        """
        self._loss = ""
        self._multi_class = None
        self.loss = loss
        self.epochs = epochs
        self.nn_batch_size = nn_batch_size
        self.validation_split = validation_split

        if callbacks:
            self.callbacks = callbacks
        else:
            self.callbacks = []

        if not metrics:
            metrics = ['mean_squared_error']

        super(KerasDnnRegressor, self).__init__(hidden_layer_sizes=hidden_layer_sizes,
                                                target_activation="linear",
                                                target_dimension=1,
                                                learning_rate=learning_rate,
                                                loss=loss,
                                                metrics=metrics,
                                                batch_normalization=batch_normalization,
                                                verbosity=verbosity,
                                                dropout_rate=dropout_rate,
                                                activations=activations,
                                                optimizer=optimizer)

    @property
    def target_activation(self):
        return "linear"

    @target_activation.setter
    def target_activation(self, value):
        if value != "linear":
            msg = "The subclass of KerasBaseRegressor does not allow to use another " \
                  "target_activation. Please use 'linear' like default."
            logger.error(msg)
            raise ValueError(msg)

    @property
    def loss(self):
        return self._loss

    @loss.setter
    def loss(self, value):
        if value in keras_dnn_base_model.get_loss_allocation()["regression"]:
            self._loss = value
        else:
            raise ValueError("Loss function is not supported. Feel free to use upperclass without restrictions.")

    def fit(self, X: np.ndarray, y: np.ndarray):
        """
        Starting the learning.

        Parameters:
            X:
                The input samples with shape [n_samples, n_features].

            y:
                The input targets with shape [n_samples, 1].

        """
        self.create_model(X.shape[1])
        super(KerasDnnBaseModel, self).fit(X, y)
        return self

__init__(self, hidden_layer_sizes=None, learning_rate=0.01, loss='mean_squared_error', epochs=10, nn_batch_size=64, metrics=None, validation_split=0.1, callbacks=None, batch_normalization=True, verbosity=0, dropout_rate=0.2, activations='relu', optimizer='adam') special

Initialize the object.

Parameters:

Name Type Description Default
hidden_layer_sizes int

Number of perceptrons per layer.

None
learning_rate float

Step size of the learning adjustment.

0.01
loss str

Loss function.

'mean_squared_error'
epochs int

Number of arbitrary cutoffs, generally defined as "one pass over the entire dataset", used to separate training into distinct phases, which is useful for logging and periodic evaluation.

10
nn_batch_size int

Typically the batch_size. A batch is a set of nn_batch_size samples. The samples in a batch are processed independently, in parallel. If training, a batch results in only one update to the model.

64
metrics list

List of evaluate metrics.

None
callbacks list

Within Keras, there is the ability to add callbacks specifically designed to be run at the end of an epoch. Examples of these are learning rate changes and model checkpointing (saving).

None
validation_split float

Split size of validation set.

0.1
batch_normalization bool

Batch normalization applies a transformation that maintains the mean output close to 0 and the output standard deviation close to 1.

True
verbosity int

The level of verbosity, 0 is least talkative and gives only warn and error, 1 gives adds info and 2 adds debug.

0
dropout_rate Union[float, list]

A Dropout layer applies random dropout and rescales the output. In inference mode, the same layer does nothing. Float -> added behind each layer List -> Same size as hidden_layer_size

0.2
activations Union[str, list]

Activation function.

'relu'
optimizer Union[tensorflow.python.keras.optimizer_v2.optimizer_v2.OptimizerV2, str]

Optimization algorithm.

'adam'
Source code in photonai/modelwrapper/keras_dnn_regressor.py
def __init__(self,
             hidden_layer_sizes: int = None,
             learning_rate: float = 0.01,
             loss: str = "mean_squared_error",
             epochs: int = 10,
             nn_batch_size: int = 64,
             metrics: list = None,
             validation_split: float = 0.1,
             callbacks: list = None,
             batch_normalization: bool = True,
             verbosity: int = 0,
             dropout_rate: Union[float, list] = 0.2,
             activations: Union[str, list] = 'relu',
             optimizer: Union[Optimizer, str] = "adam"):
    """
    Initialize the object.

    Parameters:
        hidden_layer_sizes:
            Number of perceptrons per layer.

        learning_rate:
            Step size of the learning adjustment.

        loss:
            Loss function.

        epochs:
            Number of arbitrary cutoffs, generally defined as
            "one pass over the entire dataset", used to separate training into distinct phases,
            which is useful for logging and periodic evaluation.

        nn_batch_size:
            Typically the batch_size. A batch is a set of nn_batch_size samples.
            The samples in a batch are processed independently, in parallel.
            If training, a batch results in only one update to the model.

        metrics:
            List of evaluate metrics.

        callbacks:
            Within Keras, there is the ability to add callbacks specifically designed
            to be run at the end of an epoch. Examples of these
            are learning rate changes and model checkpointing (saving).

        validation_split:
            Split size of validation set.

        batch_normalization:
            Batch normalization applies a transformation that maintains
            the mean output close to 0 and the output standard deviation close to 1.

        verbosity:
            The level of verbosity, 0 is least talkative and
            gives only warn and error, 1 gives adds info and 2 adds debug.

        dropout_rate:
            A Dropout layer applies random dropout and rescales the output.
            In inference mode, the same layer does nothing.
            Float -> added behind each layer
            List -> Same size as hidden_layer_size

        activations:
            Activation function.

        optimizer:
            Optimization algorithm.

    """
    self._loss = ""
    self._multi_class = None
    self.loss = loss
    self.epochs = epochs
    self.nn_batch_size = nn_batch_size
    self.validation_split = validation_split

    if callbacks:
        self.callbacks = callbacks
    else:
        self.callbacks = []

    if not metrics:
        metrics = ['mean_squared_error']

    super(KerasDnnRegressor, self).__init__(hidden_layer_sizes=hidden_layer_sizes,
                                            target_activation="linear",
                                            target_dimension=1,
                                            learning_rate=learning_rate,
                                            loss=loss,
                                            metrics=metrics,
                                            batch_normalization=batch_normalization,
                                            verbosity=verbosity,
                                            dropout_rate=dropout_rate,
                                            activations=activations,
                                            optimizer=optimizer)

fit(self, X, y)

Starting the learning.

Parameters:

Name Type Description Default
X ndarray

The input samples with shape [n_samples, n_features].

required
y ndarray

The input targets with shape [n_samples, 1].

required
Source code in photonai/modelwrapper/keras_dnn_regressor.py
def fit(self, X: np.ndarray, y: np.ndarray):
    """
    Starting the learning.

    Parameters:
        X:
            The input samples with shape [n_samples, n_features].

        y:
            The input targets with shape [n_samples, 1].

    """
    self.create_model(X.shape[1])
    super(KerasDnnBaseModel, self).fit(X, y)
    return self