main-letters.cpp

// #define EIGEN_USE_MKL_ALL
#include <eigen3/Eigen/Dense>
#include <iostream>
#include <memory>

#include "libs/data_loader/data_loader.h"
#include "libs/image/character.h"
#include "libs/learn/learn.h"

int main(int, char **)
{
	Eigen::initParallel();
	// extract training examples;
	std::vector<std::shared_ptr<Learn<unsigned char>::TrainingExample>> trainingExamples;
	std::vector<std::shared_ptr<Learn<unsigned char>::TrainingExample>> testingExamples;

	// DataLoader trainLoader("./data/train-images.idx3-ubyte", "./data/train-labels.idx1-ubyte");
	// DataLoader testLoader("./data/t10k-images.idx3-ubyte", "./data/t10k-labels.idx1-ubyte");

	DataLoader trainLoader("./data/emnist-balanced-train-images-idx3-ubyte",
						   "./data/emnist-balanced-train-labels-idx1-ubyte");
	DataLoader testLoader("./data/emnist-balanced-test-images-idx3-ubyte",
						  "./data/emnist-balanced-test-labels-idx1-ubyte");

	for (size_t i = 0; i < 112800; ++i) {
		trainingExamples.push_back(trainLoader.getCharacter(i));
	}

	for (size_t i = 0; i < 18800; ++i) {
		testingExamples.push_back(testLoader.getCharacter(i));
	}

	// for (size_t i = 0; i < 100; ++i)
	// 	std::cout << trainLoader.getCharacter(i)->toString() << std::endl;

	// std::cout << testLoader._numImages << std::endl;

	Learn<unsigned char> l(28 * 28, 4, 512, 62, 0.01, 0.001, 0.8, std::move(trainingExamples),
						   std::move(testingExamples), 2, 16, "save-file2.txt");

	try {
		l.loadLayerConnections();
	} catch (...) {
		l.setRandomLayerConnections();
	}
	// l.setRandomLayerConnections();
	l.train(15);

	for (size_t i = 100; i < 200; ++i)
		l.predictIndividual(testLoader.getCharacter(i));

	return 0;

	// for (size_t i = 0; i < 60000; ++i) {
	// 	trainingExamples.push_back(trainLoader.getImage(i));
	// }

	// for (size_t i = 0; i < 10000; ++i) {
	// 	testingExamples.push_back(testLoader.getImage(i));
	// }

	// Learn<double> l(28 * 28, 3, 512, 10, 0.01, 0.001, 0.8, std::move(trainingExamples), std::move(testingExamples),
	// 2, 				16, "save-file.txt");

	// try {
	// 	l.loadLayerConnections();
	// } catch (...) {
	// 	l.setRandomLayerConnections();
	// }

	// l.train(10);
	// for (size_t i = 0; i < 10; ++i)
	// 	l.predictIndividual(testLoader.getImage(i));
}