Relintai/pmlpp
1
0
Fork 0
mirror of https://github.com/Relintai/pmlpp.git synced 2025-01-30 16:49:21 +01:00
Code Issues Packages Projects Releases Wiki Activity

Initial cleanup pass on MLPPLinReg.

Browse Source
This commit is contained in:
Relintai 2023-02-11 11:09:29 +01:00
parent 2a5c278f40
commit afedf90694
3 changed files with 510 additions and 261 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

587
mlpp/lin_reg/lin_reg.cpp
View File

@ -5,6 +5,7 @@
// //
#include "lin_reg.h" #include "lin_reg.h"
#include "../cost/cost.h" #include "../cost/cost.h"
#include "../lin_alg/lin_alg.h" #include "../lin_alg/lin_alg.h"
#include "../regularization/reg.h" #include "../regularization/reg.h"
@ -15,315 +16,395 @@
#include <iostream> #include <iostream>
#include <random> #include <random>
MLPPLinReg::MLPPLinReg(std::vector<std::vector<real_t>> p_inputSet, std::vector<real_t> p_outputSet, std::string p_reg, real_t p_lambda, real_t p_alpha) { /*
inputSet = p_inputSet; Ref<MLPPMatrix> MLPPLinReg::get_input_set() {
outputSet = p_outputSet; return _input_set;
n = p_inputSet.size(); }
k = p_inputSet[0].size(); void MLPPLinReg::set_input_set(const Ref<MLPPMatrix> &val) {
reg = p_reg; _input_set = val;
lambda = p_lambda;
alpha = p_alpha;
y_hat.resize(n); _initialized = false;
weights = MLPPUtilities::weightInitialization(k);
bias = MLPPUtilities::biasInitialization();
} }
std::vector<real_t> MLPPLinReg::modelSetTest(std::vector<std::vector<real_t>> X) { Ref<MLPPVector> MLPPLinReg::get_output_set() {
return Evaluate(X); return _output_set;
}
void MLPPLinReg::set_output_set(const Ref<MLPPVector> &val) {
_output_set = val;
_initialized = false;
} }
real_t MLPPLinReg::modelTest(std::vector<real_t> x) { MLPPReg::RegularizationType MLPPLinReg::get_reg() {
return Evaluate(x); return _reg;
}
void MLPPLinReg::set_reg(const MLPPReg::RegularizationType val) {
_reg = val;
_initialized = false;
} }
void MLPPLinReg::NewtonRaphson(real_t learning_rate, int max_epoch, bool UI) { real_t MLPPLinReg::get_lambda() {
return _lambda;
}
void MLPPLinReg::set_lambda(const real_t val) {
_lambda = val;
_initialized = false;
}
real_t MLPPLinReg::get_alpha() {
return _alpha;
}
void MLPPLinReg::set_alpha(const real_t val) {
_alpha = val;
_initialized = false;
}
*/
std::vector<real_t> MLPPLinReg::model_set_test(std::vector<std::vector<real_t>> X) {
ERR_FAIL_COND_V(!_initialized, std::vector<real_t>());
return evaluatem(X);
}
real_t MLPPLinReg::model_test(std::vector<real_t> x) {
ERR_FAIL_COND_V(!_initialized, 0);
return evaluatev(x);
}
void MLPPLinReg::newton_raphson(real_t learning_rate, int max_epoch, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
forwardPass();
while (true) {
cost_prev = Cost(y_hat, outputSet);
std::vector<real_t> error = alg.subtraction(y_hat, outputSet); forward_pass();
while (true) {
cost_prev = cost(_y_hat, _output_set);
std::vector<real_t> error = alg.subtraction(_y_hat, _output_set);
// Calculating the weight gradients (2nd derivative) // Calculating the weight gradients (2nd derivative)
std::vector<real_t> first_derivative = alg.mat_vec_mult(alg.transpose(inputSet), error); std::vector<real_t> first_derivative = alg.mat_vec_mult(alg.transpose(_input_set), error);
std::vector<std::vector<real_t>> second_derivative = alg.matmult(alg.transpose(inputSet), inputSet); std::vector<std::vector<real_t>> second_derivative = alg.matmult(alg.transpose(_input_set), _input_set);
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate / n, alg.mat_vec_mult(alg.transpose(alg.inverse(second_derivative)), first_derivative))); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate / _n, alg.mat_vec_mult(alg.transpose(alg.inverse(second_derivative)), first_derivative)));
weights = regularization.regWeights(weights, lambda, alpha, reg); _weights = regularization.regWeights(_weights, _lambda, _alpha, _reg);
// Calculating the bias gradients (2nd derivative) // Calculating the bias gradients (2nd derivative)
bias -= learning_rate * alg.sum_elements(error) / n; // We keep this the same. The 2nd derivative is just [1]. _bias -= learning_rate * alg.sum_elements(error) / _n; // We keep this the same. The 2nd derivative is just [1].
forwardPass();
if (UI) { forward_pass();
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputSet));
MLPPUtilities::UI(weights, bias); if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, cost(_y_hat, _output_set));
MLPPUtilities::UI(_weights, _bias);
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
} }
void MLPPLinReg::gradientDescent(real_t learning_rate, int max_epoch, bool UI) { void MLPPLinReg::gradient_descent(real_t learning_rate, int max_epoch, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
forwardPass();
forward_pass();
while (true) { while (true) {
cost_prev = Cost(y_hat, outputSet); cost_prev = cost(_y_hat, _output_set);
std::vector<real_t> error = alg.subtraction(y_hat, outputSet); std::vector<real_t> error = alg.subtraction(_y_hat, _output_set);
// Calculating the weight gradients // Calculating the weight gradients
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate / n, alg.mat_vec_mult(alg.transpose(inputSet), error))); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate / _n, alg.mat_vec_mult(alg.transpose(_input_set), error)));
weights = regularization.regWeights(weights, lambda, alpha, reg); _weights = regularization.regWeights(_weights, _lambda, _alpha, _reg);
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / n; _bias -= learning_rate * alg.sum_elements(error) / _n;
forwardPass();
if (UI) { forward_pass();
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputSet));
MLPPUtilities::UI(weights, bias); if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, cost(_y_hat, _output_set));
MLPPUtilities::UI(_weights, _bias);
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
} }
void MLPPLinReg::SGD(real_t learning_rate, int max_epoch, bool UI) { void MLPPLinReg::sgd(real_t learning_rate, int max_epoch, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
while (true) {
std::random_device rd; std::random_device rd;
std::default_random_engine generator(rd()); std::default_random_engine generator(rd());
std::uniform_int_distribution<int> distribution(0, int(n - 1)); std::uniform_int_distribution<int> distribution(0, int(_n - 1));
while (true) {
int outputIndex = distribution(generator); int outputIndex = distribution(generator);
real_t y_hat = Evaluate(inputSet[outputIndex]); real_t y_hat = evaluatev(_input_set[outputIndex]);
cost_prev = Cost({ y_hat }, { outputSet[outputIndex] }); cost_prev = cost({ y_hat }, { _output_set[outputIndex] });
real_t error = y_hat - outputSet[outputIndex]; real_t error = y_hat - _output_set[outputIndex];
// Weight updation // Weight updation
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate * error, inputSet[outputIndex])); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate * error, _input_set[outputIndex]));
weights = regularization.regWeights(weights, lambda, alpha, reg); _weights = regularization.regWeights(_weights, _lambda, _alpha, _reg);
// Bias updation // Bias updation
bias -= learning_rate * error; _bias -= learning_rate * error;
y_hat = Evaluate({ inputSet[outputIndex] }); y_hat = evaluatev(_input_set[outputIndex]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost({ y_hat }, { outputSet[outputIndex] })); MLPPUtilities::CostInfo(epoch, cost_prev, cost({ y_hat }, { _output_set[outputIndex] }));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::MBGD(real_t learning_rate, int max_epoch, int mini_batch_size, bool UI) { void MLPPLinReg::mbgd(real_t learning_rate, int max_epoch, int mini_batch_size, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
// Creating the mini-batches // Creating the mini-batches
int n_mini_batch = n / mini_batch_size; int n_mini_batch = _n / mini_batch_size;
auto batches = MLPPUtilities::createMiniBatches(inputSet, outputSet, n_mini_batch); auto batches = MLPPUtilities::createMiniBatches(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(batches); auto inputMiniBatches = std::get<0>(batches);
auto outputMiniBatches = std::get<1>(batches); auto outputMiniBatches = std::get<1>(batches);
while (true) { while (true) {
for (int i = 0; i < n_mini_batch; i++) { for (int i = 0; i < n_mini_batch; i++) {
std::vector<real_t> y_hat = Evaluate(inputMiniBatches[i]); std::vector<real_t> y_hat = evaluatem(inputMiniBatches[i]);
cost_prev = Cost(y_hat, outputMiniBatches[i]); cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]);
// Calculating the weight gradients // Calculating the weight gradients
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error))); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error)));
weights = regularization.regWeights(weights, lambda, alpha, reg); _weights = regularization.regWeights(_weights, _lambda, _alpha, _reg);
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); _bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size();
y_hat = Evaluate(inputMiniBatches[i]); y_hat = evaluatem(inputMiniBatches[i]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i])); MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i]));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::Momentum(real_t learning_rate, int max_epoch, int mini_batch_size, real_t gamma, bool UI) { void MLPPLinReg::momentum(real_t learning_rate, int max_epoch, int mini_batch_size, real_t gamma, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
// Creating the mini-batches // Creating the mini-batches
int n_mini_batch = n / mini_batch_size; int n_mini_batch = _n / mini_batch_size;
auto batches = MLPPUtilities::createMiniBatches(inputSet, outputSet, n_mini_batch); auto batches = MLPPUtilities::createMiniBatches(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(batches); auto inputMiniBatches = std::get<0>(batches);
auto outputMiniBatches = std::get<1>(batches); auto outputMiniBatches = std::get<1>(batches);
// Initializing necessary components for Momentum. // Initializing necessary components for Momentum.
std::vector<real_t> v = alg.zerovec(weights.size()); std::vector<real_t> v = alg.zerovec(_weights.size());
while (true) { while (true) {
for (int i = 0; i < n_mini_batch; i++) { for (int i = 0; i < n_mini_batch; i++) {
std::vector<real_t> y_hat = Evaluate(inputMiniBatches[i]); std::vector<real_t> y_hat = evaluatem(inputMiniBatches[i]);
cost_prev = Cost(y_hat, outputMiniBatches[i]); cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]);
// Calculating the weight gradients // Calculating the weight gradients
std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error)); std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error));
std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(weights, lambda, alpha, reg); std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(_weights, _lambda, _alpha, _reg);
std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final
v = alg.addition(alg.scalarMultiply(gamma, v), alg.scalarMultiply(learning_rate, weight_grad)); v = alg.addition(alg.scalarMultiply(gamma, v), alg.scalarMultiply(learning_rate, weight_grad));
weights = alg.subtraction(weights, v); _weights = alg.subtraction(_weights, v);
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal _bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal
y_hat = Evaluate(inputMiniBatches[i]); y_hat = evaluatem(inputMiniBatches[i]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i])); MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i]));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::NAG(real_t learning_rate, int max_epoch, int mini_batch_size, real_t gamma, bool UI) { void MLPPLinReg::nag(real_t learning_rate, int max_epoch, int mini_batch_size, real_t gamma, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
// Creating the mini-batches // Creating the mini-batches
int n_mini_batch = n / mini_batch_size; int n_mini_batch = _n / mini_batch_size;
auto batches = MLPPUtilities::createMiniBatches(inputSet, outputSet, n_mini_batch); auto batches = MLPPUtilities::createMiniBatches(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(batches); auto inputMiniBatches = std::get<0>(batches);
auto outputMiniBatches = std::get<1>(batches); auto outputMiniBatches = std::get<1>(batches);
// Initializing necessary components for Momentum. // Initializing necessary components for Momentum.
std::vector<real_t> v = alg.zerovec(weights.size()); std::vector<real_t> v = alg.zerovec(_weights.size());
while (true) { while (true) {
for (int i = 0; i < n_mini_batch; i++) { for (int i = 0; i < n_mini_batch; i++) {
weights = alg.subtraction(weights, alg.scalarMultiply(gamma, v)); // "Aposterori" calculation _weights = alg.subtraction(_weights, alg.scalarMultiply(gamma, v)); // "Aposterori" calculation
std::vector<real_t> y_hat = Evaluate(inputMiniBatches[i]); std::vector<real_t> y_hat = evaluatem(inputMiniBatches[i]);
cost_prev = Cost(y_hat, outputMiniBatches[i]); cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]);
// Calculating the weight gradients // Calculating the weight gradients
std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error)); std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error));
std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(weights, lambda, alpha, reg); std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(_weights, _lambda, _alpha, _reg);
std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final
v = alg.addition(alg.scalarMultiply(gamma, v), alg.scalarMultiply(learning_rate, weight_grad)); v = alg.addition(alg.scalarMultiply(gamma, v), alg.scalarMultiply(learning_rate, weight_grad));
weights = alg.subtraction(weights, v); _weights = alg.subtraction(_weights, v);
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal _bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal
y_hat = Evaluate(inputMiniBatches[i]); y_hat = evaluatem(inputMiniBatches[i]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i])); MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i]));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::Adagrad(real_t learning_rate, int max_epoch, int mini_batch_size, real_t e, bool UI) { void MLPPLinReg::adagrad(real_t learning_rate, int max_epoch, int mini_batch_size, real_t e, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
// Creating the mini-batches // Creating the mini-batches
int n_mini_batch = n / mini_batch_size; int n_mini_batch = _n / mini_batch_size;
auto batches = MLPPUtilities::createMiniBatches(inputSet, outputSet, n_mini_batch); auto batches = MLPPUtilities::createMiniBatches(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(batches); auto inputMiniBatches = std::get<0>(batches);
auto outputMiniBatches = std::get<1>(batches); auto outputMiniBatches = std::get<1>(batches);
// Initializing necessary components for Adagrad. // Initializing necessary components for Adagrad.
std::vector<real_t> v = alg.zerovec(weights.size()); std::vector<real_t> v = alg.zerovec(_weights.size());
while (true) { while (true) {
for (int i = 0; i < n_mini_batch; i++) { for (int i = 0; i < n_mini_batch; i++) {
std::vector<real_t> y_hat = Evaluate(inputMiniBatches[i]); std::vector<real_t> y_hat = evaluatem(inputMiniBatches[i]);
cost_prev = Cost(y_hat, outputMiniBatches[i]); cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]);
// Calculating the weight gradients // Calculating the weight gradients
std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error)); std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error));
std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(weights, lambda, alpha, reg); std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(_weights, _lambda, _alpha, _reg);
std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final
v = alg.hadamard_product(weight_grad, weight_grad); v = alg.hadamard_product(weight_grad, weight_grad);
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(weight_grad, alg.sqrt(alg.scalarAdd(e, v))))); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(weight_grad, alg.sqrt(alg.scalarAdd(e, v)))));
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal _bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal
y_hat = Evaluate(inputMiniBatches[i]); y_hat = evaluatem(inputMiniBatches[i]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i])); MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i]));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::Adadelta(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t e, bool UI) { void MLPPLinReg::adadelta(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t e, bool ui) {
ERR_FAIL_COND(!_initialized);
// Adagrad upgrade. Momentum is applied. // Adagrad upgrade. Momentum is applied.
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
@ -331,72 +412,77 @@ void MLPPLinReg::Adadelta(real_t learning_rate, int max_epoch, int mini_batch_si
int epoch = 1; int epoch = 1;
// Creating the mini-batches // Creating the mini-batches
int n_mini_batch = n / mini_batch_size; int n_mini_batch = _n / mini_batch_size;
auto batches = MLPPUtilities::createMiniBatches(inputSet, outputSet, n_mini_batch); auto batches = MLPPUtilities::createMiniBatches(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(batches); auto inputMiniBatches = std::get<0>(batches);
auto outputMiniBatches = std::get<1>(batches); auto outputMiniBatches = std::get<1>(batches);
// Initializing necessary components for Adagrad. // Initializing necessary components for Adagrad.
std::vector<real_t> v = alg.zerovec(weights.size()); std::vector<real_t> v = alg.zerovec(_weights.size());
while (true) { while (true) {
for (int i = 0; i < n_mini_batch; i++) { for (int i = 0; i < n_mini_batch; i++) {
std::vector<real_t> y_hat = Evaluate(inputMiniBatches[i]); std::vector<real_t> y_hat = evaluatem(inputMiniBatches[i]);
cost_prev = Cost(y_hat, outputMiniBatches[i]); cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]);
// Calculating the weight gradients // Calculating the weight gradients
std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error)); std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error));
std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(weights, lambda, alpha, reg); std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(_weights, _lambda, _alpha, _reg);
std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final
v = alg.addition(alg.scalarMultiply(b1, v), alg.scalarMultiply(1 - b1, alg.hadamard_product(weight_grad, weight_grad))); v = alg.addition(alg.scalarMultiply(b1, v), alg.scalarMultiply(1 - b1, alg.hadamard_product(weight_grad, weight_grad)));
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(weight_grad, alg.sqrt(alg.scalarAdd(e, v))))); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(weight_grad, alg.sqrt(alg.scalarAdd(e, v)))));
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal _bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal
y_hat = Evaluate(inputMiniBatches[i]); y_hat = evaluatem(inputMiniBatches[i]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i])); MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i]));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::Adam(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool UI) { void MLPPLinReg::adam(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
// Creating the mini-batches // Creating the mini-batches
int n_mini_batch = n / mini_batch_size; int n_mini_batch = _n / mini_batch_size;
auto batches = MLPPUtilities::createMiniBatches(inputSet, outputSet, n_mini_batch); auto batches = MLPPUtilities::createMiniBatches(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(batches); auto inputMiniBatches = std::get<0>(batches);
auto outputMiniBatches = std::get<1>(batches); auto outputMiniBatches = std::get<1>(batches);
// Initializing necessary components for Adam. // Initializing necessary components for Adam.
std::vector<real_t> m = alg.zerovec(weights.size()); std::vector<real_t> m = alg.zerovec(_weights.size());
std::vector<real_t> v = alg.zerovec(weights.size()); std::vector<real_t> v = alg.zerovec(_weights.size());
while (true) { while (true) {
for (int i = 0; i < n_mini_batch; i++) { for (int i = 0; i < n_mini_batch; i++) {
std::vector<real_t> y_hat = Evaluate(inputMiniBatches[i]); std::vector<real_t> y_hat = evaluatem(inputMiniBatches[i]);
cost_prev = Cost(y_hat, outputMiniBatches[i]); cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]);
// Calculating the weight gradients // Calculating the weight gradients
std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error)); std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error));
std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(weights, lambda, alpha, reg); std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(_weights, _lambda, _alpha, _reg);
std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final
m = alg.addition(alg.scalarMultiply(b1, m), alg.scalarMultiply(1 - b1, weight_grad)); m = alg.addition(alg.scalarMultiply(b1, m), alg.scalarMultiply(1 - b1, weight_grad));
@ -405,50 +491,55 @@ void MLPPLinReg::Adam(real_t learning_rate, int max_epoch, int mini_batch_size,
std::vector<real_t> m_hat = alg.scalarMultiply(1 / (1 - pow(b1, epoch)), m); std::vector<real_t> m_hat = alg.scalarMultiply(1 / (1 - pow(b1, epoch)), m);
std::vector<real_t> v_hat = alg.scalarMultiply(1 / (1 - pow(b2, epoch)), v); std::vector<real_t> v_hat = alg.scalarMultiply(1 / (1 - pow(b2, epoch)), v);
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(m_hat, alg.scalarAdd(e, alg.sqrt(v_hat))))); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(m_hat, alg.scalarAdd(e, alg.sqrt(v_hat)))));
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal _bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal
y_hat = Evaluate(inputMiniBatches[i]); y_hat = evaluatem(inputMiniBatches[i]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i])); MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i]));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::Adamax(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool UI) { void MLPPLinReg::adamax(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
// Creating the mini-batches // Creating the mini-batches
int n_mini_batch = n / mini_batch_size; int n_mini_batch = _n / mini_batch_size;
auto batches = MLPPUtilities::createMiniBatches(inputSet, outputSet, n_mini_batch); auto batches = MLPPUtilities::createMiniBatches(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(batches); auto inputMiniBatches = std::get<0>(batches);
auto outputMiniBatches = std::get<1>(batches); auto outputMiniBatches = std::get<1>(batches);
std::vector<real_t> m = alg.zerovec(weights.size()); std::vector<real_t> m = alg.zerovec(_weights.size());
std::vector<real_t> u = alg.zerovec(weights.size()); std::vector<real_t> u = alg.zerovec(_weights.size());
while (true) { while (true) {
for (int i = 0; i < n_mini_batch; i++) { for (int i = 0; i < n_mini_batch; i++) {
std::vector<real_t> y_hat = Evaluate(inputMiniBatches[i]); std::vector<real_t> y_hat = evaluatem(inputMiniBatches[i]);
cost_prev = Cost(y_hat, outputMiniBatches[i]); cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]);
// Calculating the weight gradients // Calculating the weight gradients
std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error)); std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error));
std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(weights, lambda, alpha, reg); std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(_weights, _lambda, _alpha, _reg);
std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final
m = alg.addition(alg.scalarMultiply(b1, m), alg.scalarMultiply(1 - b1, weight_grad)); m = alg.addition(alg.scalarMultiply(b1, m), alg.scalarMultiply(1 - b1, weight_grad));
@ -456,51 +547,57 @@ void MLPPLinReg::Adamax(real_t learning_rate, int max_epoch, int mini_batch_size
std::vector<real_t> m_hat = alg.scalarMultiply(1 / (1 - pow(b1, epoch)), m); std::vector<real_t> m_hat = alg.scalarMultiply(1 / (1 - pow(b1, epoch)), m);
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(m_hat, u))); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(m_hat, u)));
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal _bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal
y_hat = Evaluate(inputMiniBatches[i]); y_hat = evaluatem(inputMiniBatches[i]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i])); MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i]));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::Nadam(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool UI) { void MLPPLinReg::nadam(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool ui) {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPReg regularization; MLPPReg regularization;
real_t cost_prev = 0; real_t cost_prev = 0;
int epoch = 1; int epoch = 1;
// Creating the mini-batches // Creating the mini-batches
int n_mini_batch = n / mini_batch_size; int n_mini_batch = _n / mini_batch_size;
auto batches = MLPPUtilities::createMiniBatches(inputSet, outputSet, n_mini_batch); auto batches = MLPPUtilities::createMiniBatches(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(batches); auto inputMiniBatches = std::get<0>(batches);
auto outputMiniBatches = std::get<1>(batches); auto outputMiniBatches = std::get<1>(batches);
// Initializing necessary components for Adam. // Initializing necessary components for Adam.
std::vector<real_t> m = alg.zerovec(weights.size()); std::vector<real_t> m = alg.zerovec(_weights.size());
std::vector<real_t> v = alg.zerovec(weights.size()); std::vector<real_t> v = alg.zerovec(_weights.size());
std::vector<real_t> m_final = alg.zerovec(weights.size()); std::vector<real_t> m_final = alg.zerovec(_weights.size());
while (true) { while (true) {
for (int i = 0; i < n_mini_batch; i++) { for (int i = 0; i < n_mini_batch; i++) {
std::vector<real_t> y_hat = Evaluate(inputMiniBatches[i]); std::vector<real_t> y_hat = evaluatem(inputMiniBatches[i]);
cost_prev = Cost(y_hat, outputMiniBatches[i]); cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]);
// Calculating the weight gradients // Calculating the weight gradients
std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error)); std::vector<real_t> gradient = alg.scalarMultiply(1 / outputMiniBatches[i].size(), alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), error));
std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(weights, lambda, alpha, reg); std::vector<real_t> RegDerivTerm = regularization.regDerivTerm(_weights, _lambda, _alpha, _reg);
std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final std::vector<real_t> weight_grad = alg.addition(gradient, RegDerivTerm); // Weight_grad_final
m = alg.addition(alg.scalarMultiply(b1, m), alg.scalarMultiply(1 - b1, weight_grad)); m = alg.addition(alg.scalarMultiply(b1, m), alg.scalarMultiply(1 - b1, weight_grad));
@ -510,88 +607,168 @@ void MLPPLinReg::Nadam(real_t learning_rate, int max_epoch, int mini_batch_size,
std::vector<real_t> m_hat = alg.scalarMultiply(1 / (1 - pow(b1, epoch)), m); std::vector<real_t> m_hat = alg.scalarMultiply(1 / (1 - pow(b1, epoch)), m);
std::vector<real_t> v_hat = alg.scalarMultiply(1 / (1 - pow(b2, epoch)), v); std::vector<real_t> v_hat = alg.scalarMultiply(1 / (1 - pow(b2, epoch)), v);
weights = alg.subtraction(weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(m_final, alg.scalarAdd(e, alg.sqrt(v_hat))))); _weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate, alg.elementWiseDivision(m_final, alg.scalarAdd(e, alg.sqrt(v_hat)))));
// Calculating the bias gradients // Calculating the bias gradients
bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal _bias -= learning_rate * alg.sum_elements(error) / outputMiniBatches[i].size(); // As normal
y_hat = Evaluate(inputMiniBatches[i]); y_hat = evaluatem(inputMiniBatches[i]);
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputMiniBatches[i])); MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i]));
MLPPUtilities::UI(weights, bias); MLPPUtilities::UI(_weights, _bias);
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
} }
forwardPass();
forward_pass();
} }
void MLPPLinReg::normalEquation() { void MLPPLinReg::normal_equation() {
ERR_FAIL_COND(!_initialized);
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPStat stat; MLPPStat stat;
std::vector<real_t> x_means; std::vector<real_t> x_means;
std::vector<std::vector<real_t>> inputSetT = alg.transpose(inputSet); std::vector<std::vector<real_t>> _input_setT = alg.transpose(_input_set);
x_means.resize(inputSetT.size()); x_means.resize(_input_setT.size());
for (uint32_t i = 0; i < inputSetT.size(); i++) { for (uint32_t i = 0; i < _input_setT.size(); i++) {
x_means[i] = (stat.mean(inputSetT[i])); x_means[i] = (stat.mean(_input_setT[i]));
} }
//try {
std::vector<real_t> temp; std::vector<real_t> temp;
temp.resize(k); temp.resize(_k);
temp = alg.mat_vec_mult(alg.inverse(alg.matmult(alg.transpose(inputSet), inputSet)), alg.mat_vec_mult(alg.transpose(inputSet), outputSet)); temp = alg.mat_vec_mult(alg.inverse(alg.matmult(alg.transpose(_input_set), _input_set)), alg.mat_vec_mult(alg.transpose(_input_set), _output_set));
if (std::isnan(temp[0])) {
//throw 99; ERR_FAIL_COND_MSG(std::isnan(temp[0]), "ERR: Resulting matrix was noninvertible/degenerate, and so the normal equation could not be performed. Try utilizing gradient descent.");
//TODO ERR_FAIL_COND
std::cout << "ERR: Resulting matrix was noninvertible/degenerate, and so the normal equation could not be performed. Try utilizing gradient descent." << std::endl; if (_reg == "Ridge") {
return; _weights = alg.mat_vec_mult(alg.inverse(alg.addition(alg.matmult(alg.transpose(_input_set), _input_set), alg.scalarMultiply(_lambda, alg.identity(_k)))), alg.mat_vec_mult(alg.transpose(_input_set), _output_set));
} else { } else {
if (reg == "Ridge") { _weights = alg.mat_vec_mult(alg.inverse(alg.matmult(alg.transpose(_input_set), _input_set)), alg.mat_vec_mult(alg.transpose(_input_set), _output_set));
weights = alg.mat_vec_mult(alg.inverse(alg.addition(alg.matmult(alg.transpose(inputSet), inputSet), alg.scalarMultiply(lambda, alg.identity(k)))), alg.mat_vec_mult(alg.transpose(inputSet), outputSet));
} else {
weights = alg.mat_vec_mult(alg.inverse(alg.matmult(alg.transpose(inputSet), inputSet)), alg.mat_vec_mult(alg.transpose(inputSet), outputSet));
} }
bias = stat.mean(outputSet) - alg.dot(weights, x_means); _bias = stat.mean(_output_set) - alg.dot(_weights, x_means);
forwardPass(); forward_pass();
}
//} catch (int err_num) {
// std::cout << "ERR " << err_num << ": Resulting matrix was noninvertible/degenerate, and so the normal equation could not be performed. Try utilizing gradient descent." << std::endl;
//}
} }
real_t MLPPLinReg::score() { real_t MLPPLinReg::score() {
ERR_FAIL_COND_V(!_initialized, 0);
MLPPUtilities util; MLPPUtilities util;
return util.performance(y_hat, outputSet);
return util.performance(_y_hat, _output_set);
} }
void MLPPLinReg::save(std::string fileName) { void MLPPLinReg::save(std::string fileName) {
ERR_FAIL_COND(!_initialized);
MLPPUtilities util; MLPPUtilities util;
util.saveParameters(fileName, weights, bias);
util.saveParameters(fileName, _weights, _bias);
} }
real_t MLPPLinReg::Cost(std::vector<real_t> y_hat, std::vector<real_t> y) { bool MLPPLinReg::is_initialized() {
return _initialized;
}
void MLPPLinReg::initialize() {
if (_initialized) {
return;
}
//ERR_FAIL_COND(!_input_set.is_valid() || !_output_set.is_valid());
_initialized = true;
}
MLPPLinReg::MLPPLinReg(std::vector<std::vector<real_t>> p_input_set, std::vector<real_t> p_output_set, std::string p_reg, real_t p_lambda, real_t p_alpha) {
_input_set = p_input_set;
_output_set = p_output_set;
_n = p_input_set.size();
_k = p_input_set[0].size();
_reg = p_reg;
_lambda = p_lambda;
_alpha = p_alpha;
_y_hat.resize(_n);
_weights = MLPPUtilities::weightInitialization(_k);
_bias = MLPPUtilities::biasInitialization();
_initialized = true;
}
MLPPLinReg::MLPPLinReg() {
_initialized = false;
}
MLPPLinReg::~MLPPLinReg() {
}
real_t MLPPLinReg::cost(std::vector<real_t> y_hat, std::vector<real_t> y) {
MLPPReg regularization; MLPPReg regularization;
class MLPPCost cost; MLPPCost mlpp_cost;
return cost.MSE(y_hat, y) + regularization.regTerm(weights, lambda, alpha, reg);
return mlpp_cost.MSE(y_hat, y) + regularization.regTerm(_weights, _lambda, _alpha, _reg);
} }
std::vector<real_t> MLPPLinReg::Evaluate(std::vector<std::vector<real_t>> X) { real_t MLPPLinReg::evaluatev(std::vector<real_t> x) {
MLPPLinAlg alg; MLPPLinAlg alg;
return alg.scalarAdd(bias, alg.mat_vec_mult(X, weights));
return alg.dot(_weights, x) + _bias;
} }
real_t MLPPLinReg::Evaluate(std::vector<real_t> x) { std::vector<real_t> MLPPLinReg::evaluatem(std::vector<std::vector<real_t>> X) {
MLPPLinAlg alg; MLPPLinAlg alg;
return alg.dot(weights, x) + bias;
return alg.scalarAdd(_bias, alg.mat_vec_mult(X, _weights));
} }
// wTx + b // wTx + b
void MLPPLinReg::forwardPass() { void MLPPLinReg::forward_pass() {
y_hat = Evaluate(inputSet); _y_hat = evaluatem(_input_set);
}
void MLPPLinReg::_bind_methods() {
/*
ClassDB::bind_method(D_METHOD("get_input_set"), &MLPPLinReg::get_input_set);
ClassDB::bind_method(D_METHOD("set_input_set", "val"), &MLPPLinReg::set_input_set);
ADD_PROPERTY(PropertyInfo(Variant::OBJECT, "input_set", PROPERTY_HINT_RESOURCE_TYPE, "MLPPMatrix"), "set_input_set", "get_input_set");
ClassDB::bind_method(D_METHOD("get_output_set"), &MLPPLinReg::get_output_set);
ClassDB::bind_method(D_METHOD("set_output_set", "val"), &MLPPLinReg::set_output_set);
ADD_PROPERTY(PropertyInfo(Variant::OBJECT, "output_set", PROPERTY_HINT_RESOURCE_TYPE, "MLPPVector"), "set_output_set", "get_output_set");
ClassDB::bind_method(D_METHOD("get_reg"), &MLPPLinReg::get_reg);
ClassDB::bind_method(D_METHOD("set_reg", "val"), &MLPPLinReg::set_reg);
ADD_PROPERTY(PropertyInfo(Variant::INT, "reg"), "set_reg", "get_reg");
ClassDB::bind_method(D_METHOD("get_lambda"), &MLPPLinReg::get_lambda);
ClassDB::bind_method(D_METHOD("set_lambda", "val"), &MLPPLinReg::set_lambda);
ADD_PROPERTY(PropertyInfo(Variant::REAL, "lambda"), "set_lambda", "get_lambda");
ClassDB::bind_method(D_METHOD("get_alpha"), &MLPPLinReg::get_alpha);
ClassDB::bind_method(D_METHOD("set_alpha", "val"), &MLPPLinReg::set_alpha);
ADD_PROPERTY(PropertyInfo(Variant::REAL, "alpha"), "set_alpha", "get_alpha");
ClassDB::bind_method(D_METHOD("model_test", "x"), &MLPPLinReg::model_test);
ClassDB::bind_method(D_METHOD("model_set_test", "X"), &MLPPLinReg::model_set_test);
ClassDB::bind_method(D_METHOD("gradient_descent", "learning_rate", "max_epoch", "ui"), &MLPPLinReg::gradient_descent, false);
ClassDB::bind_method(D_METHOD("sgd", "learning_rate", "max_epoch", "ui"), &MLPPLinReg::sgd, false);
ClassDB::bind_method(D_METHOD("mbgd", "learning_rate", "max_epoch", "mini_batch_size", "ui"), &MLPPLinReg::mbgd, false);
ClassDB::bind_method(D_METHOD("score"), &MLPPLinReg::score);
ClassDB::bind_method(D_METHOD("save", "file_name"), &MLPPLinReg::save);
ClassDB::bind_method(D_METHOD("is_initialized"), &MLPPLinReg::is_initialized);
ClassDB::bind_method(D_METHOD("initialize"), &MLPPLinReg::initialize);
*/
} }

102
mlpp/lin_reg/lin_reg.h
View File

@ -10,51 +10,91 @@
#include "core/math/math_defs.h" #include "core/math/math_defs.h"
#include "core/object/reference.h"
#include "../lin_alg/mlpp_matrix.h"
#include "../lin_alg/mlpp_vector.h"
#include "../regularization/reg.h"
#include <string> #include <string>
#include <vector> #include <vector>
class MLPPLinReg { class MLPPLinReg : public Reference {
GDCLASS(MLPPLinReg, Reference);
public: public:
MLPPLinReg(std::vector<std::vector<real_t>> inputSet, std::vector<real_t> outputSet, std::string reg = "None", real_t lambda = 0.5, real_t alpha = 0.5); /*
std::vector<real_t> modelSetTest(std::vector<std::vector<real_t>> X); Ref<MLPPMatrix> get_input_set();
real_t modelTest(std::vector<real_t> x); void set_input_set(const Ref<MLPPMatrix> &val);
void NewtonRaphson(real_t learning_rate, int max_epoch, bool UI);
void gradientDescent(real_t learning_rate, int max_epoch, bool UI = false);
void SGD(real_t learning_rate, int max_epoch, bool UI = false);
void Momentum(real_t learning_rate, int max_epoch, int mini_batch_size, real_t gamma, bool UI = false); Ref<MLPPVector> get_output_set();
void NAG(real_t learning_rate, int max_epoch, int mini_batch_size, real_t gamma, bool UI = false); void set_output_set(const Ref<MLPPVector> &val);
void Adagrad(real_t learning_rate, int max_epoch, int mini_batch_size, real_t e, bool UI = false);
void Adadelta(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t e, bool UI = false); MLPPReg::RegularizationType get_reg();
void Adam(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool UI = false); void set_reg(const MLPPReg::RegularizationType val);
void Adamax(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool UI = false);
void Nadam(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool UI = false); real_t get_lambda();
void set_lambda(const real_t val);
real_t get_alpha();
void set_alpha(const real_t val);
*/
std::vector<real_t> model_set_test(std::vector<std::vector<real_t>> X);
real_t model_test(std::vector<real_t> x);
void newton_raphson(real_t learning_rate, int max_epoch, bool ui = false);
void gradient_descent(real_t learning_rate, int max_epoch, bool ui = false);
void sgd(real_t learning_rate, int max_epoch, bool ui = false);
void momentum(real_t learning_rate, int max_epoch, int mini_batch_size, real_t gamma, bool ui = false);
void nag(real_t learning_rate, int max_epoch, int mini_batch_size, real_t gamma, bool ui = false);
void adagrad(real_t learning_rate, int max_epoch, int mini_batch_size, real_t e, bool ui = false);
void adadelta(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t e, bool ui = false);
void adam(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool ui = false);
void adamax(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool ui = false);
void nadam(real_t learning_rate, int max_epoch, int mini_batch_size, real_t b1, real_t b2, real_t e, bool ui = false);
void mbgd(real_t learning_rate, int max_epoch, int mini_batch_size, bool ui = false);
void normal_equation();
void MBGD(real_t learning_rate, int max_epoch, int mini_batch_size, bool UI = false);
void normalEquation();
real_t score(); real_t score();
void save(std::string fileName); void save(std::string fileName);
private: bool is_initialized();
real_t Cost(std::vector<real_t> y_hat, std::vector<real_t> y); void initialize();
std::vector<real_t> Evaluate(std::vector<std::vector<real_t>> X); MLPPLinReg(std::vector<std::vector<real_t>> p_input_set, std::vector<real_t> p_output_set, std::string p_reg = "None", real_t p_lambda = 0.5, real_t p_alpha = 0.5);
real_t Evaluate(std::vector<real_t> x);
void forwardPass();
std::vector<std::vector<real_t>> inputSet; MLPPLinReg();
std::vector<real_t> outputSet; ~MLPPLinReg();
std::vector<real_t> y_hat;
std::vector<real_t> weights;
real_t bias;
int n; protected:
int k; real_t cost(std::vector<real_t> y_hat, std::vector<real_t> y);
real_t evaluatev(std::vector<real_t> x);
std::vector<real_t> evaluatem(std::vector<std::vector<real_t>> X);
void forward_pass();
static void _bind_methods();
std::vector<std::vector<real_t>> _input_set;
std::vector<real_t> _output_set;
std::vector<real_t> _y_hat;
std::vector<real_t> _weights;
real_t _bias;
int _n;
int _k;
// Regularization Params // Regularization Params
std::string reg; std::string _reg;
int lambda; int _lambda;
int alpha; /* This is the controlling param for Elastic Net*/ int _alpha; /* This is the controlling param for Elastic Net*/
bool _initialized;
}; };
#endif /* LinReg_hpp */ #endif /* LinReg_hpp */

76
test/mlpp_tests.cpp
View File

@ -243,10 +243,13 @@ void MLPPTests::test_multivariate_linear_regression_gradient_descent(bool ui) {
Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path); Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path);
MLPPLinReg model(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg MLPPLinRegOld model_old(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
model_old.gradientDescent(0.001, 30, ui);
alg.printVector(model_old.modelSetTest(ds->get_input()->to_std_vector()));
model.gradientDescent(0.001, 30, ui); MLPPLinReg model(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
alg.printVector(model.modelSetTest(ds->get_input()->to_std_vector())); model.gradient_descent(0.001, 30, ui);
alg.printVector(model.model_set_test(ds->get_input()->to_std_vector()));
} }
void MLPPTests::test_multivariate_linear_regression_sgd(bool ui) { void MLPPTests::test_multivariate_linear_regression_sgd(bool ui) {
@ -255,10 +258,13 @@ void MLPPTests::test_multivariate_linear_regression_sgd(bool ui) {
Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path); Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path);
MLPPLinReg model(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg MLPPLinRegOld model_old(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
model_old.SGD(0.00000001, 300000, ui);
alg.printVector(model_old.modelSetTest(ds->get_input()->to_std_vector()));
model.SGD(0.00000001, 300000, ui); MLPPLinReg model(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
alg.printVector(model.modelSetTest(ds->get_input()->to_std_vector())); model.sgd(0.00000001, 300000, ui);
alg.printVector(model.model_set_test(ds->get_input()->to_std_vector()));
} }
void MLPPTests::test_multivariate_linear_regression_mbgd(bool ui) { void MLPPTests::test_multivariate_linear_regression_mbgd(bool ui) {
@ -267,10 +273,13 @@ void MLPPTests::test_multivariate_linear_regression_mbgd(bool ui) {
Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path); Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path);
MLPPLinReg model(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg MLPPLinRegOld model_old(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
model_old.MBGD(0.001, 10000, 2, ui);
alg.printVector(model_old.modelSetTest(ds->get_input()->to_std_vector()));
model.MBGD(0.001, 10000, 2, ui); MLPPLinReg model(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
alg.printVector(model.modelSetTest(ds->get_input()->to_std_vector())); model.mbgd(0.001, 10000, 2, ui);
alg.printVector(model.model_set_test(ds->get_input()->to_std_vector()));
} }
void MLPPTests::test_multivariate_linear_regression_normal_equation(bool ui) { void MLPPTests::test_multivariate_linear_regression_normal_equation(bool ui) {
@ -279,10 +288,13 @@ void MLPPTests::test_multivariate_linear_regression_normal_equation(bool ui) {
Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path); Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path);
MLPPLinReg model(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg MLPPLinRegOld model_old(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
model_old.normalEquation();
alg.printVector(model_old.modelSetTest(ds->get_input()->to_std_vector()));
model.normalEquation(); MLPPLinReg model(ds->get_input()->to_std_vector(), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
alg.printVector(model.modelSetTest(ds->get_input()->to_std_vector())); model.normal_equation();
alg.printVector(model.model_set_test(ds->get_input()->to_std_vector()));
} }
void MLPPTests::test_multivariate_linear_regression_adam() { void MLPPTests::test_multivariate_linear_regression_adam() {
@ -291,9 +303,13 @@ void MLPPTests::test_multivariate_linear_regression_adam() {
Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path); Ref<MLPPDataSimple> ds = data.load_california_housing(_california_housing_data_path);
MLPPLinReg adamModel(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector()); MLPPLinRegOld adamModelOld(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
alg.printVector(adamModel.modelSetTest(ds->get_input()->to_std_vector())); alg.printVector(adamModelOld.modelSetTest(ds->get_input()->to_std_vector()));
std::cout << "ACCURACY: " << 100 * adamModel.score() << "%" << std::endl; std::cout << "ACCURACY: " << 100 * adamModelOld.score() << "%" << std::endl;
MLPPLinReg adam_model(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
alg.printVector(adam_model.model_set_test(ds->get_input()->to_std_vector()));
std::cout << "ACCURACY: " << 100 * adam_model.score() << "%" << std::endl;
} }
void MLPPTests::test_multivariate_linear_regression_score_sgd_adam(bool ui) { void MLPPTests::test_multivariate_linear_regression_score_sgd_adam(bool ui) {
@ -307,12 +323,20 @@ void MLPPTests::test_multivariate_linear_regression_score_sgd_adam(bool ui) {
real_t scoreSGD = 0; real_t scoreSGD = 0;
real_t scoreADAM = 0; real_t scoreADAM = 0;
for (int i = 0; i < TRIAL_NUM; i++) { for (int i = 0; i < TRIAL_NUM; i++) {
MLPPLinRegOld modelf_old(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
modelf_old.MBGD(0.001, 5, 1, ui);
scoreSGD += modelf_old.score();
MLPPLinReg modelf(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector()); MLPPLinReg modelf(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
modelf.MBGD(0.001, 5, 1, ui); modelf.mbgd(0.001, 5, 1, ui);
scoreSGD += modelf.score(); scoreSGD += modelf.score();
MLPPLinRegOld adamModelf_old(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
adamModelf_old.Adam(0.1, 5, 1, 0.9, 0.999, 1e-8, ui); // Change batch size = sgd, bgd
scoreADAM += adamModelf_old.score();
MLPPLinReg adamModelf(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector()); MLPPLinReg adamModelf(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
adamModelf.Adam(0.1, 5, 1, 0.9, 0.999, 1e-8, ui); // Change batch size = sgd, bgd adamModelf.adam(0.1, 5, 1, 0.9, 0.999, 1e-8, ui); // Change batch size = sgd, bgd
scoreADAM += adamModelf.score(); scoreADAM += adamModelf.score();
} }
@ -330,9 +354,13 @@ void MLPPTests::test_multivariate_linear_regression_epochs_gradient_descent(bool
std::cout << "Total epoch num: 300" << std::endl; std::cout << "Total epoch num: 300" << std::endl;
std::cout << "Method: 1st Order w/ Jacobians" << std::endl; std::cout << "Method: 1st Order w/ Jacobians" << std::endl;
MLPPLinRegOld model3_old(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
model3_old.gradientDescent(0.001, 300, ui);
alg.printVector(model3_old.modelSetTest(ds->get_input()->to_std_vector()));
MLPPLinReg model3(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg MLPPLinReg model3(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector()); // Can use Lasso, Ridge, ElasticNet Reg
model3.gradientDescent(0.001, 300, ui); model3.gradient_descent(0.001, 300, ui);
alg.printVector(model3.modelSetTest(ds->get_input()->to_std_vector())); alg.printVector(model3.model_set_test(ds->get_input()->to_std_vector()));
} }
void MLPPTests::test_multivariate_linear_regression_newton_raphson(bool ui) { void MLPPTests::test_multivariate_linear_regression_newton_raphson(bool ui) {
@ -344,10 +372,14 @@ void MLPPTests::test_multivariate_linear_regression_newton_raphson(bool ui) {
std::cout << "--------------------------------------------" << std::endl; std::cout << "--------------------------------------------" << std::endl;
std::cout << "Total epoch num: 300" << std::endl; std::cout << "Total epoch num: 300" << std::endl;
std::cout << "Method: Newtonian 2nd Order w/ Hessians" << std::endl; std::cout << "Method: Newtonian 2nd Order w/ Hessians" << std::endl;
MLPPLinReg model2(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
model2.NewtonRaphson(1.5, 300, ui); MLPPLinRegOld model2_old(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
alg.printVector(model2.modelSetTest(ds->get_input()->to_std_vector())); model2_old.NewtonRaphson(1.5, 300, ui);
alg.printVector(model2_old.modelSetTest(ds->get_input()->to_std_vector()));
MLPPLinReg model2(alg.transpose(ds->get_input()->to_std_vector()), ds->get_output()->to_std_vector());
model2.newton_raphson(1.5, 300, ui);
alg.printVector(model2.model_set_test(ds->get_input()->to_std_vector()));
} }
void MLPPTests::test_logistic_regression(bool ui) { void MLPPTests::test_logistic_regression(bool ui) {