Relintai/pmlpp
1
0
Fork 0
mirror of https://github.com/Relintai/pmlpp.git synced 2024-12-22 15:06:47 +01:00
Code Issues Packages Projects Releases Wiki Activity

Now MLPPTanhReg uses engine classes.

Browse Source
This commit is contained in:
Relintai 2023-02-13 21:10:19 +01:00
parent d6037730b5
commit aa8043621e
2 changed files with 104 additions and 70 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

143
mlpp/tanh_reg/tanh_reg.cpp
View File

@ -12,7 +12,6 @@
#include "../regularization/reg.h" #include "../regularization/reg.h"
#include "../utilities/utilities.h" #include "../utilities/utilities.h"
#include <iostream>
#include <random> #include <random>
/* /*
@ -62,11 +61,14 @@ void MLPPTanhReg::set_alpha(const real_t val) {
} }
*/ */
std::vector<real_t> MLPPTanhReg::model_set_test(std::vector<std::vector<real_t>> X) { // Ref<MLPPVector> model_set_test(const Ref<MLPPMatrix> &X);
// real_t model_test(const Ref<MLPPVector> &x);
Ref<MLPPVector> MLPPTanhReg::model_set_test(const Ref<MLPPMatrix> &X) {
return evaluatem(X); return evaluatem(X);
} }
real_t MLPPTanhReg::model_test(std::vector<real_t> x) { real_t MLPPTanhReg::model_test(const Ref<MLPPVector> &x) {
return evaluatev(x); return evaluatev(x);
} }
@ -83,21 +85,21 @@ void MLPPTanhReg::gradient_descent(real_t learning_rate, int max_epoch, bool ui)
while (true) { while (true) {
cost_prev = cost(_y_hat, _output_set); cost_prev = cost(_y_hat, _output_set);
std::vector<real_t> error = alg.subtraction(_y_hat, _output_set); Ref<MLPPVector> error = alg.subtractionnv(_y_hat, _output_set);
_weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate / _n, alg.mat_vec_mult(alg.transpose(_input_set), alg.hadamard_product(error, avn.tanh(_z, 1))))); _weights = alg.subtractionnv(_weights, alg.scalar_multiplynv(learning_rate / _n, alg.mat_vec_multv(alg.transposem(_input_set), alg.hadamard_productnv(error, avn.tanh_derivv(_z)))));
//_reg //_reg
_weights = regularization.regWeights(_weights, _lambda, _alpha, "None"); _weights = regularization.reg_weightsv(_weights, _lambda, _alpha, MLPPReg::REGULARIZATION_TYPE_NONE);
// Calculating the bias gradients // Calculating the bias gradients
_bias -= learning_rate * alg.sum_elements(alg.hadamard_product(error, avn.tanh(_z, 1))) / _n; _bias -= learning_rate * alg.sum_elementsv(alg.hadamard_productnv(error, avn.tanh_derivv(_z))) / _n;
forward_pass(); forward_pass();
// UI PORTION // UI PORTION
if (ui) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, cost(_y_hat, _output_set)); MLPPUtilities::cost_info(epoch, cost_prev, cost(_y_hat, _output_set));
MLPPUtilities::UI(_weights, _bias); MLPPUtilities::print_ui_vb(_weights, _bias);
} }
epoch++; epoch++;
@ -119,28 +121,47 @@ void MLPPTanhReg::sgd(real_t learning_rate, int max_epoch, bool ui) {
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));
Ref<MLPPVector> input_set_row_tmp;
input_set_row_tmp.instance();
input_set_row_tmp->resize(_input_set->size().x);
Ref<MLPPVector> output_set_row_tmp;
output_set_row_tmp.instance();
output_set_row_tmp->resize(1);
Ref<MLPPVector> y_hat_row_tmp;
y_hat_row_tmp.instance();
y_hat_row_tmp->resize(1);
while (true) { while (true) {
int outputIndex = distribution(generator); int output_index = distribution(generator);
real_t y_hat = evaluatev(_input_set[outputIndex]); _input_set->get_row_into_mlpp_vector(output_index, input_set_row_tmp);
cost_prev = cost({ _y_hat }, { _output_set[outputIndex] }); real_t output_set_entry = _output_set->get_element(output_index);
output_set_row_tmp->set_element(0, output_set_entry);
real_t error = y_hat - _output_set[outputIndex]; real_t y_hat = evaluatev(input_set_row_tmp);
y_hat_row_tmp->set_element(0, y_hat);
cost_prev = cost(y_hat_row_tmp, output_set_row_tmp);
real_t error = y_hat - output_set_entry;
// Weight Updation // Weight Updation
_weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate * error * (1 - y_hat * y_hat), _input_set[outputIndex])); _weights = alg.subtractionnv(_weights, alg.scalar_multiplynv(learning_rate * error * (1 - y_hat * y_hat), input_set_row_tmp));
//_reg //_reg
_weights = regularization.regWeights(_weights, _lambda, _alpha, "None"); _weights = regularization.reg_weightsv(_weights, _lambda, _alpha, MLPPReg::REGULARIZATION_TYPE_NONE);
// Bias updation // Bias updation
_bias -= learning_rate * error * (1 - y_hat * y_hat); _bias -= learning_rate * error * (1 - y_hat * y_hat);
y_hat = evaluatev(_input_set[outputIndex]); y_hat = evaluatev(input_set_row_tmp);
if (ui) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, cost({ _y_hat }, { _output_set[outputIndex] })); MLPPUtilities::cost_info(epoch, cost_prev, cost(y_hat_row_tmp, output_set_row_tmp));
MLPPUtilities::UI(_weights, _bias); MLPPUtilities::print_ui_vb(_weights, _bias);
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
@ -161,33 +182,34 @@ void MLPPTanhReg::mbgd(real_t learning_rate, int max_epoch, int mini_batch_size,
// 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(_input_set, _output_set, n_mini_batch); MLPPUtilities::CreateMiniBatchMVBatch batches = MLPPUtilities::create_mini_batchesmv(_input_set, _output_set, n_mini_batch);
auto inputMiniBatches = std::get<0>(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 = evaluatem(inputMiniBatches[i]); Ref<MLPPMatrix> current_input_batch_entry = batches.input_sets[i];
std::vector<real_t> z = propagatem(inputMiniBatches[i]); Ref<MLPPVector> current_output_batch_entry = batches.output_sets[i];
cost_prev = cost(y_hat, outputMiniBatches[i]);
std::vector<real_t> error = alg.subtraction(y_hat, outputMiniBatches[i]); Ref<MLPPVector> y_hat = evaluatem(current_input_batch_entry);
Ref<MLPPVector> z = propagatem(current_input_batch_entry);
cost_prev = cost(y_hat, current_output_batch_entry);
Ref<MLPPVector> error = alg.subtractionnv(y_hat, current_output_batch_entry);
// Calculating the weight gradients // Calculating the weight gradients
_weights = alg.subtraction(_weights, alg.scalarMultiply(learning_rate / _n, alg.mat_vec_mult(alg.transpose(inputMiniBatches[i]), alg.hadamard_product(error, avn.tanh(z, 1))))); _weights = alg.subtractionnv(_weights, alg.scalar_multiplynv(learning_rate / _n, alg.mat_vec_multv(alg.transposem(current_input_batch_entry), alg.hadamard_productnv(error, avn.tanh_derivv(z)))));
//_reg //_reg
_weights = regularization.regWeights(_weights, _lambda, _alpha, "None"); _weights = regularization.reg_weightsv(_weights, _lambda, _alpha, MLPPReg::REGULARIZATION_TYPE_NONE);
// Calculating the bias gradients // Calculating the bias gradients
_bias -= learning_rate * alg.sum_elements(alg.hadamard_product(error, avn.tanh(_z, true))) / _n; _bias -= learning_rate * alg.sum_elementsv(alg.hadamard_productnv(error, avn.tanh_derivv(_z))) / _n;
forward_pass(); forward_pass();
y_hat = evaluatem(inputMiniBatches[i]); y_hat = evaluatem(current_input_batch_entry);
if (ui) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, cost(y_hat, outputMiniBatches[i])); MLPPUtilities::cost_info(epoch, cost_prev, cost(y_hat, current_output_batch_entry));
MLPPUtilities::UI(_weights, _bias); MLPPUtilities::print_ui_vb(_weights, _bias);
} }
} }
@ -204,13 +226,13 @@ void MLPPTanhReg::mbgd(real_t learning_rate, int max_epoch, int mini_batch_size,
real_t MLPPTanhReg::score() { real_t MLPPTanhReg::score() {
MLPPUtilities util; MLPPUtilities util;
return util.performance(_y_hat, _output_set); return util.performance_vec(_y_hat, _output_set);
} }
void MLPPTanhReg::save(std::string file_name) { void MLPPTanhReg::save(const String &file_name) {
MLPPUtilities util; //MLPPUtilities util;
util.saveParameters(file_name, _weights, _bias); //util.saveParameters(file_name, _weights, _bias);
} }
bool MLPPTanhReg::is_initialized() { bool MLPPTanhReg::is_initialized() {
@ -226,53 +248,68 @@ void MLPPTanhReg::initialize() {
_initialized = true; _initialized = true;
} }
MLPPTanhReg::MLPPTanhReg(std::vector<std::vector<real_t>> p_input_set, std::vector<real_t> p_output_set, MLPPReg::RegularizationType p_reg, real_t p_lambda, real_t p_alpha) { MLPPTanhReg::MLPPTanhReg(const Ref<MLPPMatrix> &p_input_set, const Ref<MLPPVector> &p_output_set, MLPPReg::RegularizationType p_reg, real_t p_lambda, real_t p_alpha) {
_input_set = p_input_set; _input_set = p_input_set;
_output_set = p_output_set; _output_set = p_output_set;
_n = _input_set.size(); _n = _input_set->size().y;
_k = _input_set[0].size(); _k = _input_set->size().x;
_reg = p_reg; _reg = p_reg;
_lambda = p_lambda; _lambda = p_lambda;
_alpha = p_alpha; _alpha = p_alpha;
_y_hat.resize(_n); _y_hat.instance();
_weights = MLPPUtilities::weightInitialization(_k); _y_hat->resize(_n);
_bias = MLPPUtilities::biasInitialization();
MLPPUtilities utils;
_weights.instance();
_weights->resize(_k);
utils.weight_initializationv(_weights);
_bias = utils.bias_initializationr();
_initialized = true;
} }
MLPPTanhReg::MLPPTanhReg() { MLPPTanhReg::MLPPTanhReg() {
_initialized = false;
} }
MLPPTanhReg::~MLPPTanhReg() { MLPPTanhReg::~MLPPTanhReg() {
} }
real_t MLPPTanhReg::cost(std::vector<real_t> y_hat, std::vector<real_t> y) { real_t MLPPTanhReg::cost(const Ref<MLPPVector> &y_hat, const Ref<MLPPVector> &y) {
MLPPReg regularization; MLPPReg regularization;
class MLPPCost cost; MLPPCost mlpp_cost;
//_reg //_reg
return cost.MSE(y_hat, y) + regularization.regTerm(_weights, _lambda, _alpha, "None"); return mlpp_cost.msev(y_hat, y) + regularization.reg_termv(_weights, _lambda, _alpha, MLPPReg::REGULARIZATION_TYPE_NONE);
} }
real_t MLPPTanhReg::evaluatev(std::vector<real_t> x) { real_t MLPPTanhReg::evaluatev(const Ref<MLPPVector> &x) {
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPActivation avn; MLPPActivation avn;
return avn.tanh(alg.dot(_weights, x) + _bias);
return avn.tanh_normr(alg.dotv(_weights, x) + _bias);
} }
real_t MLPPTanhReg::propagatev(std::vector<real_t> x) { real_t MLPPTanhReg::propagatev(const Ref<MLPPVector> &x) {
MLPPLinAlg alg; MLPPLinAlg alg;
return alg.dot(_weights, x) + _bias;
return alg.dotv(_weights, x) + _bias;
} }
std::vector<real_t> MLPPTanhReg::evaluatem(std::vector<std::vector<real_t>> X) { Ref<MLPPVector> MLPPTanhReg::evaluatem(const Ref<MLPPMatrix> &X) {
MLPPLinAlg alg; MLPPLinAlg alg;
MLPPActivation avn; MLPPActivation avn;
return avn.tanh(alg.scalarAdd(_bias, alg.mat_vec_mult(X, _weights)));
return avn.tanh_normv(alg.scalar_addnv(_bias, alg.mat_vec_multv(X, _weights)));
} }
std::vector<real_t> MLPPTanhReg::propagatem(std::vector<std::vector<real_t>> X) { Ref<MLPPVector> MLPPTanhReg::propagatem(const Ref<MLPPMatrix> &X) {
MLPPLinAlg alg; MLPPLinAlg alg;
return alg.scalarAdd(_bias, alg.mat_vec_mult(X, _weights));
return alg.scalar_addnv(_bias, alg.mat_vec_multv(X, _weights));
} }
// Tanh ( wTx + b ) // Tanh ( wTx + b )
@ -280,7 +317,7 @@ void MLPPTanhReg::forward_pass() {
MLPPActivation avn; MLPPActivation avn;
_z = propagatem(_input_set); _z = propagatem(_input_set);
_y_hat = avn.tanh(_z); _y_hat = avn.tanh_normv(_z);
} }
void MLPPTanhReg::_bind_methods() { void MLPPTanhReg::_bind_methods() {

31
mlpp/tanh_reg/tanh_reg.h
View File

@ -17,9 +17,6 @@
#include "../regularization/reg.h" #include "../regularization/reg.h"
#include <string>
#include <vector>
class MLPPTanhReg : public Reference { class MLPPTanhReg : public Reference {
GDCLASS(MLPPTanhReg, Reference); GDCLASS(MLPPTanhReg, Reference);
@ -41,8 +38,8 @@ public:
void set_alpha(const real_t val); void set_alpha(const real_t val);
*/ */
std::vector<real_t> model_set_test(std::vector<std::vector<real_t>> X); Ref<MLPPVector> model_set_test(const Ref<MLPPMatrix> &X);
real_t model_test(std::vector<real_t> x); real_t model_test(const Ref<MLPPVector> &x);
void gradient_descent(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 sgd(real_t learning_rate, int max_epoch, bool ui = false);
@ -50,34 +47,34 @@ public:
real_t score(); real_t score();
void save(std::string file_name); void save(const String &file_name);
bool is_initialized(); bool is_initialized();
void initialize(); void initialize();
MLPPTanhReg(std::vector<std::vector<real_t>> p_input_set, std::vector<real_t> p_output_set, MLPPReg::RegularizationType p_reg = MLPPReg::REGULARIZATION_TYPE_NONE, real_t p_lambda = 0.5, real_t p_alpha = 0.5); MLPPTanhReg(const Ref<MLPPMatrix> &p_input_set, const Ref<MLPPVector> &p_output_set, MLPPReg::RegularizationType p_reg = MLPPReg::REGULARIZATION_TYPE_NONE, real_t p_lambda = 0.5, real_t p_alpha = 0.5);
MLPPTanhReg(); MLPPTanhReg();
~MLPPTanhReg(); ~MLPPTanhReg();
protected: protected:
real_t cost(std::vector<real_t> y_hat, std::vector<real_t> y); real_t cost(const Ref<MLPPVector> &y_hat, const Ref<MLPPVector> &y);
real_t evaluatev(std::vector<real_t> x); real_t evaluatev(const Ref<MLPPVector> &x);
real_t propagatev(std::vector<real_t> x); real_t propagatev(const Ref<MLPPVector> &x);
std::vector<real_t> evaluatem(std::vector<std::vector<real_t>> X); Ref<MLPPVector> evaluatem(const Ref<MLPPMatrix> &X);
std::vector<real_t> propagatem(std::vector<std::vector<real_t>> X); Ref<MLPPVector> propagatem(const Ref<MLPPMatrix> &X);
void forward_pass(); void forward_pass();
static void _bind_methods(); static void _bind_methods();
std::vector<std::vector<real_t>> _input_set; Ref<MLPPMatrix> _input_set;
std::vector<real_t> _output_set; Ref<MLPPVector> _output_set;
std::vector<real_t> _z; Ref<MLPPVector> _z;
std::vector<real_t> _y_hat; Ref<MLPPVector> _y_hat;
std::vector<real_t> _weights; Ref<MLPPVector> _weights;
real_t _bias; real_t _bias;
int _n; int _n;