arbitragelab.ml_approach.neural_networks

This module implements the Multi Layer Perceptron, RNN model and the Pi Sigma Model described in Dunis et al. (2005).

Module Contents

Classes

BaseNeuralNetwork	Skeleton Class to be inherited by child neural network implementations.
MultiLayerPerceptron	Vanilla Multi Layer Perceptron implementation.
RecurrentNeuralNetwork	Recurrent Neural Network implementation.
PiSigmaNeuralNetwork	Pi Sigma Neural Network implementation.

class BaseNeuralNetwork

Skeleton Class to be inherited by child neural network implementations.

fit(*args, **kwargs)

Wrapper over the keras model fit function.

Returns:

(History) Fitted model.

predict(*args, **kwargs)

Wrapper over the keras model predict function.

plot_loss()

Method that returns visual plot of the loss trajectory in terms of epochs spent training.

class MultiLayerPerceptron(frame_size: int, hidden_size: int = 2, num_outputs: int = 1, loss_fn: str = 'mean_squared_error', optmizer: str = 'adam', metrics: str = 'accuracy', hidden_layer_activation_function: str = 'relu', output_layer_act_func: str = 'linear')

Bases: BaseNeuralNetwork

Vanilla Multi Layer Perceptron implementation.

build()

Builds and compiles model architecture.

Returns:

(Model) Resulting model.

fit(*args, **kwargs)

Wrapper over the keras model fit function.

Returns:

(History) Fitted model.

predict(*args, **kwargs)

Wrapper over the keras model predict function.

plot_loss()

Method that returns visual plot of the loss trajectory in terms of epochs spent training.

class RecurrentNeuralNetwork(input_shape: tuple, hidden_size: int = 10, num_outputs: int = 1, loss_fn: str = 'mean_squared_error', optmizer: str = 'adam', metrics: str = 'accuracy', hidden_layer_activation_function: str = 'relu', output_layer_act_func: str = 'linear')

Bases: BaseNeuralNetwork

Recurrent Neural Network implementation.

build()

Builds and compiles model architecture.

Returns:

(Model) Resulting model.

fit(*args, **kwargs)

Wrapper over the keras model fit function.

Returns:

(History) Fitted model.

predict(*args, **kwargs)

Wrapper over the keras model predict function.

plot_loss()

Method that returns visual plot of the loss trajectory in terms of epochs spent training.

class PiSigmaNeuralNetwork(frame_size: int, hidden_size: int = 2, num_outputs: int = 1, loss_fn: str = 'mean_squared_error', optmizer: str = 'sgd', metrics: str = 'accuracy', hidden_layer_activation_function: str = 'linear', output_layer_act_func: str = 'sigmoid')

Bases: BaseNeuralNetwork

Pi Sigma Neural Network implementation.

build()

Builds and compiles model architecture.

Returns:

(Model) Resulting model.

fit(*args, **kwargs)

Wrapper over the keras model fit function.

Returns:

(History) Fitted model.

predict(*args, **kwargs)

Wrapper over the keras model predict function.

plot_loss()

Method that returns visual plot of the loss trajectory in terms of epochs spent training.