pmlpp/mlpp/softmax_reg/softmax_reg.cpp

//
//  SoftmaxReg.cpp
//
//  Created by Marc Melikyan on 10/2/20.
//

#include "softmax_reg.h"
#include "../activation/activation.h"
#include "../cost/cost.h"
#include "../lin_alg/lin_alg.h"
#include "../regularization/reg.h"
#include "../utilities/utilities.h"

#include <iostream>
#include <random>

namespace MLPP {
SoftmaxReg::SoftmaxReg(std::vector<std::vector<double>> inputSet, std::vector<std::vector<double>> outputSet, std::string reg, double lambda, double alpha) :
		inputSet(inputSet), outputSet(outputSet), n(inputSet.size()), k(inputSet[0].size()), n_class(outputSet[0].size()), reg(reg), lambda(lambda), alpha(alpha) {
	y_hat.resize(n);
	weights = Utilities::weightInitialization(k, n_class);
	bias = Utilities::biasInitialization(n_class);
}

std::vector<double> SoftmaxReg::modelTest(std::vector<double> x) {
	return Evaluate(x);
}

std::vector<std::vector<double>> SoftmaxReg::modelSetTest(std::vector<std::vector<double>> X) {
	return Evaluate(X);
}

void SoftmaxReg::gradientDescent(double learning_rate, int max_epoch, bool UI) {
	LinAlg alg;
	Reg regularization;
	double cost_prev = 0;
	int epoch = 1;
	forwardPass();

	while (true) {
		cost_prev = Cost(y_hat, outputSet);
		std::vector<std::vector<double>> error = alg.subtraction(y_hat, outputSet);

		//Calculating the weight gradients
		std::vector<std::vector<double>> w_gradient = alg.matmult(alg.transpose(inputSet), error);

		//Weight updation
		weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate, w_gradient));
		weights = regularization.regWeights(weights, lambda, alpha, reg);

		// Calculating the bias gradients
		//double b_gradient = alg.sum_elements(error);

		// Bias Updation
		bias = alg.subtractMatrixRows(bias, alg.scalarMultiply(learning_rate, error));

		forwardPass();

		// UI PORTION
		if (UI) {
			Utilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputSet));
			Utilities::UI(weights, bias);
		}
		epoch++;

		if (epoch > max_epoch) {
			break;
		}
	}
}

void SoftmaxReg::SGD(double learning_rate, int max_epoch, bool UI) {
	LinAlg alg;
	Reg regularization;
	double cost_prev = 0;
	int epoch = 1;

	while (true) {
		std::random_device rd;
		std::default_random_engine generator(rd());
		std::uniform_int_distribution<int> distribution(0, int(n - 1));
		double outputIndex = distribution(generator);

		std::vector<double> y_hat = Evaluate(inputSet[outputIndex]);
		cost_prev = Cost({ y_hat }, { outputSet[outputIndex] });

		// Calculating the weight gradients
		std::vector<std::vector<double>> w_gradient = alg.outerProduct(inputSet[outputIndex], alg.subtraction(y_hat, outputSet[outputIndex]));

		// Weight Updation
		weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate, w_gradient));
		weights = regularization.regWeights(weights, lambda, alpha, reg);

		// Calculating the bias gradients
		std::vector<double> b_gradient = alg.subtraction(y_hat, outputSet[outputIndex]);

		// Bias updation
		bias = alg.subtraction(bias, alg.scalarMultiply(learning_rate, b_gradient));

		y_hat = Evaluate({ inputSet[outputIndex] });

		if (UI) {
			Utilities::CostInfo(epoch, cost_prev, Cost({ y_hat }, { outputSet[outputIndex] }));
			Utilities::UI(weights, bias);
		}
		epoch++;

		if (epoch > max_epoch) {
			break;
		}
	}
	forwardPass();
}

void SoftmaxReg::MBGD(double learning_rate, int max_epoch, int mini_batch_size, bool UI) {
	LinAlg alg;
	Reg regularization;
	double cost_prev = 0;
	int epoch = 1;

	// Creating the mini-batches
	int n_mini_batch = n / mini_batch_size;
	auto [inputMiniBatches, outputMiniBatches] = Utilities::createMiniBatches(inputSet, outputSet, n_mini_batch);

	while (true) {
		for (int i = 0; i < n_mini_batch; i++) {
			std::vector<std::vector<double>> y_hat = Evaluate(inputMiniBatches[i]);
			cost_prev = Cost(y_hat, outputMiniBatches[i]);

			std::vector<std::vector<double>> error = alg.subtraction(y_hat, outputMiniBatches[i]);

			// Calculating the weight gradients
			std::vector<std::vector<double>> w_gradient = alg.matmult(alg.transpose(inputMiniBatches[i]), error);

			//Weight updation
			weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate, w_gradient));
			weights = regularization.regWeights(weights, lambda, alpha, reg);

			// Calculating the bias gradients
			bias = alg.subtractMatrixRows(bias, alg.scalarMultiply(learning_rate, error));
			y_hat = Evaluate(inputMiniBatches[i]);

			if (UI) {
				Utilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i]));
				Utilities::UI(weights, bias);
			}
		}
		epoch++;
		if (epoch > max_epoch) {
			break;
		}
	}
	forwardPass();
}

double SoftmaxReg::score() {
	Utilities util;
	return util.performance(y_hat, outputSet);
}

void SoftmaxReg::save(std::string fileName) {
	Utilities util;
	util.saveParameters(fileName, weights, bias);
}

double SoftmaxReg::Cost(std::vector<std::vector<double>> y_hat, std::vector<std::vector<double>> y) {
	Reg regularization;
	class Cost cost;
	return cost.CrossEntropy(y_hat, y) + regularization.regTerm(weights, lambda, alpha, reg);
}

std::vector<double> SoftmaxReg::Evaluate(std::vector<double> x) {
	LinAlg alg;
	Activation avn;
	return avn.softmax(alg.addition(bias, alg.mat_vec_mult(alg.transpose(weights), x)));
}

std::vector<std::vector<double>> SoftmaxReg::Evaluate(std::vector<std::vector<double>> X) {
	LinAlg alg;
	Activation avn;

	return avn.softmax(alg.mat_vec_add(alg.matmult(X, weights), bias));
}

// softmax ( wTx + b )
void SoftmaxReg::forwardPass() {
	LinAlg alg;
	Activation avn;

	y_hat = avn.softmax(alg.mat_vec_add(alg.matmult(inputSet, weights), bias));
}
} //namespace MLPP