magmadnn::model::NeuralNetwork< T > Class Template Reference

Inheritance diagram for magmadnn::model::NeuralNetwork< T >:

Collaboration diagram for magmadnn::model::NeuralNetwork< T >:

Public Member Functions
	NeuralNetwork (std::vector< layer::Layer< T > *> layers, optimizer::loss_t loss_func, optimizer::optimizer_t optimizer, nn_params_t params)
	NeuralNetwork (std::vector< layer::Layer< T > *> layers, optimizer::loss_t loss_func, optimizer::Optimizer< T > *optim, nn_params_t params)
virtual magmadnn_error_t	fit (Tensor< T > *x, Tensor< T > *y, metric_t &metric_out, bool verbose=false)
virtual Tensor< T > *	predict (Tensor< T > *sample)
virtual unsigned int	predict_class (Tensor< T > *sample)
virtual std::vector< layer::Layer< T > * >	get_layers ()
Public Member Functions inherited from magmadnn::model::Model< T >
	Model ()
virtual double	get_accuracy ()
virtual double	get_loss ()
virtual double	get_training_time ()
virtual std::string	get_name ()

Protected Attributes
std::vector< layer::Layer< T > * >	layers
optimizer::loss_t	loss_func
optimizer::optimizer_t	optimizer
nn_params_t	model_params
op::Operation< T > *	network_input_op_ptr
op::Operation< T > *	network_output_op_ptr
Tensor< T > *	network_input_tensor_ptr
Tensor< T > *	network_output_tensor_ptr
op::Operation< T > *	ground_truth_op_ptr
Tensor< T > *	ground_truth_tensor_ptr
T	default_learning_rate = (T) 0.05
std::vector< op::Operation< T > * >	_vars
op::Operation< T > *	_obj
Tensor< T > *	_obj_tensor_ptr
optimizer::Optimizer< T > *	optim
Protected Attributes inherited from magmadnn::model::Model< T >
std::string	_name
metric_t	_last_training_metric

Constructor & Destructor Documentation

NeuralNetwork() [1/2]

template<typename T >

magmadnn::model::NeuralNetwork< T >::NeuralNetwork	(	std::vector< layer::Layer< T > *>	layers,
		optimizer::loss_t	loss_func,
		optimizer::optimizer_t	optimizer,
		nn_params_t	params
	)

Constructs a neural network with the given parameters.

Parameters

layers	a vector of the layers comprising the network.
loss_func	a loss_t enumerant describing the loss function to use at the end of the network.
optimizer	a optimizer_t enumerant describing the optimizer to use in training.
params	a set of neural network parameters.

NeuralNetwork() [2/2]

template<typename T >

magmadnn::model::NeuralNetwork< T >::NeuralNetwork	(	std::vector< layer::Layer< T > *>	layers,
		optimizer::loss_t	loss_func,
		optimizer::Optimizer< T > *	optim,
		nn_params_t	params
	)

Constructs a neural network. Allows you to supply custom optimizer.

Parameters

layers	a vector of the layers comprising the network.
loss_func	a loss_t enumerant describing the loss function to use at the end of the network.
optim	Optimizer<T> that will be used to optimize the network
params	a set of neural network parameters.

Member Function Documentation

fit()

template<typename T >

magmadnn_error_t magmadnn::model::NeuralNetwork< T >::fit ( Tensor< T > *  x, Tensor< T > *  y, metric_t &  metric_out, bool  verbose = false  )

virtual

Trains this neural network using x and y. The batch size and epochs are used from the model parameters given during construction. If verbose, then training info is printed periodically.

Parameters

x	independent data
y	ground truth one-hot encoded data
metric_out	[out] metric to write training metrics in
verbose	if true, run in verbose mode

Returns

magmadnn_error_t 0 on success

Implements magmadnn::model::Model< T >.

predict()

template<typename T >

Tensor< T > * magmadnn::model::NeuralNetwork< T >::predict ( Tensor< T > *  sample )

virtual

return a one-hot encoded output of the network on this sample

Parameters

sample

a single sample

Returns

Tensor<T>* one-hot encoded prediction

Implements magmadnn::model::Model< T >.

Here is the call graph for this function:

predict_class()

template<typename T >

unsigned int magmadnn::model::NeuralNetwork< T >::predict_class ( Tensor< T > *  sample )

virtual

This is equivalent to math::argmax(predict(sample)), where the one-hot encoded predict value is used.

Parameters

sample

Returns

unsigned int the index of the predicted class.

Implements magmadnn::model::Model< T >.

---

The documentation for this class was generated from the following files:

• include/model/neuralnetwork/neuralnetwork.h
• src/model/neuralnetwork/neuralnetwork.cpp