Class implementing a simple MNIST patch softmax classification with multi-layer neural net - imported from previously loaded auto-encoder net and adds softmax layer "at the top". More...

#include <MNISTPatchSoftmaxApplication.hpp>

Inheritance diagram for mic::applications::MNISTPatchSoftmaxApplication:

[legend]

Collaboration diagram for mic::applications::MNISTPatchSoftmaxApplication:

[legend]

Public Member Functions
	MNISTPatchSoftmaxApplication (std::string node_name_="mnist_patch_autoencoder_reconstruction")

virtual	~MNISTPatchSoftmaxApplication ()

Protected Member Functions
virtual void	initializePropertyDependentVariables ()

virtual void	initialize (int argc, char *argv[])

virtual bool	performLearningStep ()

virtual void	collectTestStatistics ()

virtual void	populateTestStatistics ()

Private Attributes
mic::importers::MNISTPatchImporter *	training_dataset_importer
	Importer responsible for loading training dataset. More...

mic::importers::MNISTPatchImporter *	test_dataset_importer
	Importer responsible for loading testing dataset. More...

mic::encoders::MatrixXfMatrixXfEncoder *	mnist_encoder
	MNIST matrix encoder. More...

mic::encoders::UIntMatrixXfEncoder *	label_encoder
	Label 2 matrix encoder (1 hot). More...

WindowMatrix2D *	w2d_input
	Window for displaying the input image. More...

WindowCollectorChart< float > *	w_chart
	Window for displaying chart with statistics. More...

WindowProbability *	w_prob
	Window displaying the probabilities. More...

mic::types::MatrixXfPtr	input_image
	Input image/matrix. More...

mic::types::MatrixXfPtr	decoded_prediction
	Probability. More...

mic::types::MatrixXfPtr	input_target
	Probability. More...

size_t	patch_size
	Size of the patch - copied from importers. More...

mic::utils::DataCollectorPtr < std::string, float >	collector_ptr
	Data collector. More...

mic::configuration::Property < std::string >	autoencoder_filename
	Property: name of the file to which the original auto-encoder will be loaded from. More...

mic::configuration::Property < size_t >	autoencoder_layers_to_be_removed
	Property: number of layers to be removed (starting from the last one). More...

mic::configuration::Property < std::string >	softmax_filename
	Property: name of the file with softmax nn. More...

mic::configuration::Property < bool >	softmax_save
	Property: flag denoting whether the nn should be saved to a file (after every episode end). More...

mic::configuration::Property < bool >	softmax_load
	Property: flag denoting whether the nn should be loaded from a file (at the initialization of the task) - if not, the program will try to import and reshape the autoencoder. More...

BackpropagationNeuralNetwork < float >	neural_net
	Multi-layer neural network. More...

Detailed Description

Class implementing a simple MNIST patch softmax classification with multi-layer neural net - imported from previously loaded auto-encoder net and adds softmax layer "at the top".

Author: tkornuta

Definition at line 52 of file MNISTPatchSoftmaxApplication.hpp.

Constructor & Destructor Documentation

mic::applications::MNISTPatchSoftmaxApplication::MNISTPatchSoftmaxApplication ( std::string node_name_ = "mnist_patch_autoencoder_reconstruction" )

Default Constructor. Sets the application/node name, default values of variables, initializes classifier etc.

Parameters

node_name_ Name of the application/node (in configuration file).

Definition at line 58 of file MNISTPatchSoftmaxApplication.cpp.

References autoencoder_filename, autoencoder_layers_to_be_removed, softmax_filename, softmax_load, softmax_save, test_dataset_importer, and training_dataset_importer.

mic::applications::MNISTPatchSoftmaxApplication::~MNISTPatchSoftmaxApplication ( )

virtual

Destructor.

Definition at line 79 of file MNISTPatchSoftmaxApplication.cpp.

References label_encoder, mnist_encoder, test_dataset_importer, training_dataset_importer, w2d_input, w_chart, and w_prob.

Member Function Documentation

void mic::applications::MNISTPatchSoftmaxApplication::collectTestStatistics ( )

protectedvirtual

Collects test statistics, executed in every testing step.

Definition at line 191 of file MNISTPatchSoftmaxApplication.cpp.

References collector_ptr, label_encoder, mnist_encoder, neural_net, mic::mlnn::BackpropagationNeuralNetwork< eT >::test(), and test_dataset_importer.

Here is the call graph for this function:

void mic::applications::MNISTPatchSoftmaxApplication::initialize	(	int	argc,
		char *	argv[]
	)

protectedvirtual

Method initializes GLUT and OpenGL windows.

Parameters

argc	Number of application parameters.
argv	Array of application parameters.

Definition at line 91 of file MNISTPatchSoftmaxApplication.cpp.

References collector_ptr, w2d_input, w_chart, and w_prob.

void mic::applications::MNISTPatchSoftmaxApplication::initializePropertyDependentVariables ( )

protectedvirtual

Initializes all variables that are property-dependent, imports MNIST datasets, for both learning and testing purposes.

Definition at line 112 of file MNISTPatchSoftmaxApplication.cpp.

References autoencoder_filename, autoencoder_layers_to_be_removed, decoded_prediction, input_image, input_target, label_encoder, mic::mlnn::MultiLayerNeuralNetwork< eT >::load(), mnist_encoder, neural_net, patch_size, mic::mlnn::MultiLayerNeuralNetwork< eT >::popLayer(), mic::mlnn::MultiLayerNeuralNetwork< eT >::pushLayer(), mic::mlnn::BackpropagationNeuralNetwork< eT >::setLoss(), softmax_filename, softmax_load, test_dataset_importer, training_dataset_importer, w2d_input, and w_prob.

Here is the call graph for this function:

bool mic::applications::MNISTPatchSoftmaxApplication::performLearningStep ( )

protectedvirtual

Performs learning step.

Definition at line 164 of file MNISTPatchSoftmaxApplication.cpp.

References collector_ptr, mic::mlnn::MultiLayerNeuralNetwork< eT >::getPredictions(), label_encoder, mnist_encoder, neural_net, mic::mlnn::BackpropagationNeuralNetwork< eT >::train(), and training_dataset_importer.

Here is the call graph for this function:

void mic::applications::MNISTPatchSoftmaxApplication::populateTestStatistics ( )

protectedvirtual

Populates test statistics, executed only every number_of_averaged_test_measures testing steps.

Definition at line 210 of file MNISTPatchSoftmaxApplication.cpp.

References neural_net, mic::mlnn::MultiLayerNeuralNetwork< eT >::save(), softmax_filename, and softmax_save.

Here is the call graph for this function:

Member Data Documentation

mic::configuration::Property<std::string> mic::applications::MNISTPatchSoftmaxApplication::autoencoder_filename

private

Property: name of the file to which the original auto-encoder will be loaded from.

Definition at line 132 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by initializePropertyDependentVariables(), and MNISTPatchSoftmaxApplication().

mic::configuration::Property<size_t> mic::applications::MNISTPatchSoftmaxApplication::autoencoder_layers_to_be_removed

private

Property: number of layers to be removed (starting from the last one).

Definition at line 135 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by initializePropertyDependentVariables(), and MNISTPatchSoftmaxApplication().

mic::utils::DataCollectorPtr<std::string, float> mic::applications::MNISTPatchSoftmaxApplication::collector_ptr

private

Data collector.

Definition at line 129 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by collectTestStatistics(), initialize(), and performLearningStep().

mic::types::MatrixXfPtr mic::applications::MNISTPatchSoftmaxApplication::decoded_prediction

private

Probability.

Definition at line 120 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by initializePropertyDependentVariables().

mic::types::MatrixXfPtr mic::applications::MNISTPatchSoftmaxApplication::input_image

private

Input image/matrix.

Definition at line 117 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by initializePropertyDependentVariables().

mic::types::MatrixXfPtr mic::applications::MNISTPatchSoftmaxApplication::input_target

private

Probability.

Definition at line 123 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by initializePropertyDependentVariables().

mic::encoders::UIntMatrixXfEncoder* mic::applications::MNISTPatchSoftmaxApplication::label_encoder

private

Label 2 matrix encoder (1 hot).

Definition at line 105 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by collectTestStatistics(), initializePropertyDependentVariables(), performLearningStep(), and ~MNISTPatchSoftmaxApplication().

mic::encoders::MatrixXfMatrixXfEncoder* mic::applications::MNISTPatchSoftmaxApplication::mnist_encoder

private

MNIST matrix encoder.

Definition at line 102 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by collectTestStatistics(), initializePropertyDependentVariables(), performLearningStep(), and ~MNISTPatchSoftmaxApplication().

BackpropagationNeuralNetwork<float> mic::applications::MNISTPatchSoftmaxApplication::neural_net

private

Multi-layer neural network.

Definition at line 147 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by collectTestStatistics(), initializePropertyDependentVariables(), performLearningStep(), and populateTestStatistics().

size_t mic::applications::MNISTPatchSoftmaxApplication::patch_size

private

Size of the patch - copied from importers.

Definition at line 126 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by initializePropertyDependentVariables().

mic::configuration::Property<std::string> mic::applications::MNISTPatchSoftmaxApplication::softmax_filename

private

Property: name of the file with softmax nn.

Definition at line 138 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by initializePropertyDependentVariables(), MNISTPatchSoftmaxApplication(), and populateTestStatistics().

mic::configuration::Property<bool> mic::applications::MNISTPatchSoftmaxApplication::softmax_load

private

Property: flag denoting whether the nn should be loaded from a file (at the initialization of the task) - if not, the program will try to import and reshape the autoencoder.

Definition at line 144 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by initializePropertyDependentVariables(), and MNISTPatchSoftmaxApplication().

mic::configuration::Property<bool> mic::applications::MNISTPatchSoftmaxApplication::softmax_save

private

Property: flag denoting whether the nn should be saved to a file (after every episode end).

Definition at line 141 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by MNISTPatchSoftmaxApplication(), and populateTestStatistics().

mic::importers::MNISTPatchImporter* mic::applications::MNISTPatchSoftmaxApplication::test_dataset_importer

private

Importer responsible for loading testing dataset.

Definition at line 99 of file MNISTPatchSoftmaxApplication.hpp.

Referenced by collectTestStatistics(), initializePropertyDependentVariables(), MNISTPatchSoftmaxApplication(), and ~MNISTPatchSoftmaxApplication().