Relintai/pmlpp
1
0
Fork 0
mirror of https://github.com/Relintai/pmlpp.git synced 2025-04-24 04:33:21 +02:00
Code Issues Packages Projects Releases Wiki Activity

Initial cleanup pass on MLPPMANN.

Browse Source
This commit is contained in:
Relintai 2023-02-11 09:53:16 +01:00
parent 7bc5a5bc1d
commit f5bd46c211
4 changed files with 239 additions and 123 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

261
mlpp/mann/mann.cpp
View File

@ -13,106 +13,128 @@
#include <iostream> #include <iostream>
MLPPMANN::MLPPMANN(std::vector<std::vector<real_t>> inputSet, std::vector<std::vector<real_t>> outputSet) : /*
inputSet(inputSet), outputSet(outputSet), n(inputSet.size()), k(inputSet[0].size()), n_output(outputSet[0].size()) { Ref<MLPPMatrix> MLPPMANN::get_input_set() {
return input_set;
}
void MLPPMANN::set_input_set(const Ref<MLPPMatrix> &val) {
input_set = val;
_initialized = false;
} }
MLPPMANN::~MLPPMANN() { Ref<MLPPMatrix> MLPPMANN::get_output_set() {
delete outputLayer; return output_set;
} }
void MLPPMANN::set_output_set(const Ref<MLPPMatrix> &val) {
output_set = val;
std::vector<std::vector<real_t>> MLPPMANN::modelSetTest(std::vector<std::vector<real_t>> X) { _initialized = false;
if (!network.empty()) { }
network[0].input = X; */
network[0].forwardPass();
for (uint32_t i = 1; i < network.size(); i++) { std::vector<std::vector<real_t>> MLPPMANN::model_set_test(std::vector<std::vector<real_t>> X) {
network[i].input = network[i - 1].a; ERR_FAIL_COND_V(!_initialized, std::vector<std::vector<real_t>>());
network[i].forwardPass();
if (!_network.empty()) {
_network[0].input = X;
_network[0].forwardPass();
for (uint32_t i = 1; i < _network.size(); i++) {
_network[i].input = _network[i - 1].a;
_network[i].forwardPass();
} }
outputLayer->input = network[network.size() - 1].a; _output_layer->input = _network[_network.size() - 1].a;
} else { } else {
outputLayer->input = X; _output_layer->input = X;
} }
outputLayer->forwardPass();
return outputLayer->a; _output_layer->forwardPass();
return _output_layer->a;
} }
std::vector<real_t> MLPPMANN::modelTest(std::vector<real_t> x) { std::vector<real_t> MLPPMANN::model_test(std::vector<real_t> x) {
if (!network.empty()) { ERR_FAIL_COND_V(!_initialized, std::vector<real_t>());
network[0].Test(x);
for (uint32_t i = 1; i < network.size(); i++) { if (!_network.empty()) {
network[i].Test(network[i - 1].a_test); _network[0].Test(x);
for (uint32_t i = 1; i < _network.size(); i++) {
_network[i].Test(_network[i - 1].a_test);
} }
outputLayer->Test(network[network.size() - 1].a_test); _output_layer->Test(_network[_network.size() - 1].a_test);
} else { } else {
outputLayer->Test(x); _output_layer->Test(x);
} }
return outputLayer->a_test; return _output_layer->a_test;
} }
void MLPPMANN::gradientDescent(real_t learning_rate, int max_epoch, bool UI) { void MLPPMANN::gradient_descent(real_t learning_rate, int max_epoch, bool ui) {
class MLPPCost cost; ERR_FAIL_COND(!_initialized);
MLPPCost mlpp_cost;
MLPPActivation avn; MLPPActivation avn;
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);
if (outputLayer->activation == "Softmax") { if (_output_layer->activation == "Softmax") {
outputLayer->delta = alg.subtraction(y_hat, outputSet); _output_layer->delta = alg.subtraction(_y_hat, _output_set);
} else { } else {
auto costDeriv = outputLayer->costDeriv_map[outputLayer->cost]; auto costDeriv = _output_layer->costDeriv_map[_output_layer->cost];
auto outputAvn = outputLayer->activation_map[outputLayer->activation]; auto outputAvn = _output_layer->activation_map[_output_layer->activation];
outputLayer->delta = alg.hadamard_product((cost.*costDeriv)(y_hat, outputSet), (avn.*outputAvn)(outputLayer->z, 1)); _output_layer->delta = alg.hadamard_product((mlpp_cost.*costDeriv)(_y_hat, _output_set), (avn.*outputAvn)(_output_layer->z, 1));
} }
std::vector<std::vector<real_t>> outputWGrad = alg.matmult(alg.transpose(outputLayer->input), outputLayer->delta); std::vector<std::vector<real_t>> outputWGrad = alg.matmult(alg.transpose(_output_layer->input), _output_layer->delta);
outputLayer->weights = alg.subtraction(outputLayer->weights, alg.scalarMultiply(learning_rate / n, outputWGrad)); _output_layer->weights = alg.subtraction(_output_layer->weights, alg.scalarMultiply(learning_rate / _n, outputWGrad));
outputLayer->weights = regularization.regWeights(outputLayer->weights, outputLayer->lambda, outputLayer->alpha, outputLayer->reg); _output_layer->weights = regularization.regWeights(_output_layer->weights, _output_layer->lambda, _output_layer->alpha, _output_layer->reg);
outputLayer->bias = alg.subtractMatrixRows(outputLayer->bias, alg.scalarMultiply(learning_rate / n, outputLayer->delta)); _output_layer->bias = alg.subtractMatrixRows(_output_layer->bias, alg.scalarMultiply(learning_rate / _n, _output_layer->delta));
if (!network.empty()) { if (!_network.empty()) {
auto hiddenLayerAvn = network[network.size() - 1].activation_map[network[network.size() - 1].activation]; auto hiddenLayerAvn = _network[_network.size() - 1].activation_map[_network[_network.size() - 1].activation];
network[network.size() - 1].delta = alg.hadamard_product(alg.matmult(outputLayer->delta, alg.transpose(outputLayer->weights)), (avn.*hiddenLayerAvn)(network[network.size() - 1].z, 1)); _network[_network.size() - 1].delta = alg.hadamard_product(alg.matmult(_output_layer->delta, alg.transpose(_output_layer->weights)), (avn.*hiddenLayerAvn)(_network[_network.size() - 1].z, true));
std::vector<std::vector<real_t>> hiddenLayerWGrad = alg.matmult(alg.transpose(network[network.size() - 1].input), network[network.size() - 1].delta); std::vector<std::vector<real_t>> hiddenLayerWGrad = alg.matmult(alg.transpose(_network[_network.size() - 1].input), _network[_network.size() - 1].delta);
network[network.size() - 1].weights = alg.subtraction(network[network.size() - 1].weights, alg.scalarMultiply(learning_rate / n, hiddenLayerWGrad)); _network[_network.size() - 1].weights = alg.subtraction(_network[_network.size() - 1].weights, alg.scalarMultiply(learning_rate / _n, hiddenLayerWGrad));
network[network.size() - 1].weights = regularization.regWeights(network[network.size() - 1].weights, network[network.size() - 1].lambda, network[network.size() - 1].alpha, network[network.size() - 1].reg); _network[_network.size() - 1].weights = regularization.regWeights(_network[_network.size() - 1].weights, _network[_network.size() - 1].lambda, _network[_network.size() - 1].alpha, _network[_network.size() - 1].reg);
network[network.size() - 1].bias = alg.subtractMatrixRows(network[network.size() - 1].bias, alg.scalarMultiply(learning_rate / n, network[network.size() - 1].delta)); _network[_network.size() - 1].bias = alg.subtractMatrixRows(_network[_network.size() - 1].bias, alg.scalarMultiply(learning_rate / _n, _network[_network.size() - 1].delta));
for (int i = network.size() - 2; i >= 0; i--) { for (int i = _network.size() - 2; i >= 0; i--) {
hiddenLayerAvn = network[i].activation_map[network[i].activation]; hiddenLayerAvn = _network[i].activation_map[_network[i].activation];
network[i].delta = alg.hadamard_product(alg.matmult(network[i + 1].delta, network[i + 1].weights), (avn.*hiddenLayerAvn)(network[i].z, 1)); _network[i].delta = alg.hadamard_product(alg.matmult(_network[i + 1].delta, _network[i + 1].weights), (avn.*hiddenLayerAvn)(_network[i].z, true));
hiddenLayerWGrad = alg.matmult(alg.transpose(network[i].input), network[i].delta); hiddenLayerWGrad = alg.matmult(alg.transpose(_network[i].input), _network[i].delta);
network[i].weights = alg.subtraction(network[i].weights, alg.scalarMultiply(learning_rate / n, hiddenLayerWGrad)); _network[i].weights = alg.subtraction(_network[i].weights, alg.scalarMultiply(learning_rate / _n, hiddenLayerWGrad));
network[i].weights = regularization.regWeights(network[i].weights, network[i].lambda, network[i].alpha, network[i].reg); _network[i].weights = regularization.regWeights(_network[i].weights, _network[i].lambda, _network[i].alpha, _network[i].reg);
network[i].bias = alg.subtractMatrixRows(network[i].bias, alg.scalarMultiply(learning_rate / n, network[i].delta)); _network[i].bias = alg.subtractMatrixRows(_network[i].bias, alg.scalarMultiply(learning_rate / _n, _network[i].delta));
} }
} }
forwardPass(); forward_pass();
if (UI) { if (ui) {
MLPPUtilities::CostInfo(epoch, cost_prev, Cost(y_hat, outputSet)); MLPPUtilities::CostInfo(epoch, cost_prev, cost(_y_hat, _output_set));
std::cout << "Layer " << network.size() + 1 << ": " << std::endl; std::cout << "Layer " << _network.size() + 1 << ": " << std::endl;
MLPPUtilities::UI(outputLayer->weights, outputLayer->bias); MLPPUtilities::UI(_output_layer->weights, _output_layer->bias);
if (!network.empty()) { if (!_network.empty()) {
std::cout << "Layer " << network.size() << ": " << std::endl; std::cout << "Layer " << _network.size() << ": " << std::endl;
for (int i = network.size() - 1; i >= 0; i--) { for (int i = _network.size() - 1; i >= 0; i--) {
std::cout << "Layer " << i + 1 << ": " << std::endl; std::cout << "Layer " << i + 1 << ": " << std::endl;
MLPPUtilities::UI(network[i].weights, network[i].bias); MLPPUtilities::UI(_network[i].weights, _network[i].bias);
} }
} }
} }
epoch++; epoch++;
if (epoch > max_epoch) { if (epoch > max_epoch) {
break; break;
} }
@ -120,69 +142,120 @@ void MLPPMANN::gradientDescent(real_t learning_rate, int max_epoch, bool UI) {
} }
real_t MLPPMANN::score() { real_t MLPPMANN::score() {
ERR_FAIL_COND_V(!_initialized, 0);
MLPPUtilities util; MLPPUtilities util;
forwardPass();
return util.performance(y_hat, outputSet); forward_pass();
return util.performance(_y_hat, _output_set);
} }
void MLPPMANN::save(std::string fileName) { void MLPPMANN::save(std::string fileName) {
ERR_FAIL_COND(!_initialized);
MLPPUtilities util; MLPPUtilities util;
if (!network.empty()) { if (!_network.empty()) {
util.saveParameters(fileName, network[0].weights, network[0].bias, 0, 1); util.saveParameters(fileName, _network[0].weights, _network[0].bias, false, 1);
for (uint32_t i = 1; i < network.size(); i++) { for (uint32_t i = 1; i < _network.size(); i++) {
util.saveParameters(fileName, network[i].weights, network[i].bias, 1, i + 1); util.saveParameters(fileName, _network[i].weights, _network[i].bias, true, i + 1);
} }
util.saveParameters(fileName, outputLayer->weights, outputLayer->bias, 1, network.size() + 1); util.saveParameters(fileName, _output_layer->weights, _output_layer->bias, true, _network.size() + 1);
} else { } else {
util.saveParameters(fileName, outputLayer->weights, outputLayer->bias, 0, network.size() + 1); util.saveParameters(fileName, _output_layer->weights, _output_layer->bias, false, _network.size() + 1);
} }
} }
void MLPPMANN::addLayer(int n_hidden, std::string activation, std::string weightInit, std::string reg, real_t lambda, real_t alpha) { void MLPPMANN::add_layer(int n_hidden, std::string activation, std::string weightInit, std::string reg, real_t lambda, real_t alpha) {
if (network.empty()) { if (_network.empty()) {
network.push_back(MLPPOldHiddenLayer(n_hidden, activation, inputSet, weightInit, reg, lambda, alpha)); _network.push_back(MLPPOldHiddenLayer(n_hidden, activation, _input_set, weightInit, reg, lambda, alpha));
network[0].forwardPass(); _network[0].forwardPass();
} else { } else {
network.push_back(MLPPOldHiddenLayer(n_hidden, activation, network[network.size() - 1].a, weightInit, reg, lambda, alpha)); _network.push_back(MLPPOldHiddenLayer(n_hidden, activation, _network[_network.size() - 1].a, weightInit, reg, lambda, alpha));
network[network.size() - 1].forwardPass(); _network[_network.size() - 1].forwardPass();
} }
} }
void MLPPMANN::addOutputLayer(std::string activation, std::string loss, std::string weightInit, std::string reg, real_t lambda, real_t alpha) { void MLPPMANN::add_output_layer(std::string activation, std::string loss, std::string weightInit, std::string reg, real_t lambda, real_t alpha) {
if (!network.empty()) { if (!_network.empty()) {
outputLayer = new MLPPOldMultiOutputLayer(n_output, network[0].n_hidden, activation, loss, network[network.size() - 1].a, weightInit, reg, lambda, alpha); _output_layer = new MLPPOldMultiOutputLayer(_n_output, _network[0].n_hidden, activation, loss, _network[_network.size() - 1].a, weightInit, reg, lambda, alpha);
} else { } else {
outputLayer = new MLPPOldMultiOutputLayer(n_output, k, activation, loss, inputSet, weightInit, reg, lambda, alpha); _output_layer = new MLPPOldMultiOutputLayer(_n_output, _k, activation, loss, _input_set, weightInit, reg, lambda, alpha);
} }
} }
real_t MLPPMANN::Cost(std::vector<std::vector<real_t>> y_hat, std::vector<std::vector<real_t>> y) { bool MLPPMANN::is_initialized() {
return _initialized;
}
void MLPPMANN::initialize() {
if (_initialized) {
return;
}
//ERR_FAIL_COND(!input_set.is_valid() || !output_set.is_valid() || n_hidden == 0);
_initialized = true;
}
MLPPMANN::MLPPMANN(std::vector<std::vector<real_t>> p_input_set, std::vector<std::vector<real_t>> p_output_set) {
_input_set = p_input_set;
_output_set = p_output_set;
_n = _input_set.size();
_k = _input_set[0].size();
_n_output = _output_set[0].size();
_initialized = true;
}
MLPPMANN::MLPPMANN() {
_initialized = false;
}
MLPPMANN::~MLPPMANN() {
delete _output_layer;
}
real_t MLPPMANN::cost(std::vector<std::vector<real_t>> y_hat, std::vector<std::vector<real_t>> y) {
MLPPReg regularization; MLPPReg regularization;
class MLPPCost cost; class MLPPCost cost;
real_t totalRegTerm = 0; real_t totalRegTerm = 0;
auto cost_function = outputLayer->cost_map[outputLayer->cost]; auto cost_function = _output_layer->cost_map[_output_layer->cost];
if (!network.empty()) { if (!_network.empty()) {
for (uint32_t i = 0; i < network.size() - 1; i++) { for (uint32_t i = 0; i < _network.size() - 1; i++) {
totalRegTerm += regularization.regTerm(network[i].weights, network[i].lambda, network[i].alpha, network[i].reg); totalRegTerm += regularization.regTerm(_network[i].weights, _network[i].lambda, _network[i].alpha, _network[i].reg);
} }
} }
return (cost.*cost_function)(y_hat, y) + totalRegTerm + regularization.regTerm(outputLayer->weights, outputLayer->lambda, outputLayer->alpha, outputLayer->reg); return (cost.*cost_function)(y_hat, y) + totalRegTerm + regularization.regTerm(_output_layer->weights, _output_layer->lambda, _output_layer->alpha, _output_layer->reg);
} }
void MLPPMANN::forwardPass() { void MLPPMANN::forward_pass() {
if (!network.empty()) { if (!_network.empty()) {
network[0].input = inputSet; _network[0].input = _input_set;
network[0].forwardPass(); _network[0].forwardPass();
for (uint32_t i = 1; i < network.size(); i++) { for (uint32_t i = 1; i < _network.size(); i++) {
network[i].input = network[i - 1].a; _network[i].input = _network[i - 1].a;
network[i].forwardPass(); _network[i].forwardPass();
} }
outputLayer->input = network[network.size() - 1].a; _output_layer->input = _network[_network.size() - 1].a;
} else { } else {
outputLayer->input = inputSet; _output_layer->input = _input_set;
} }
outputLayer->forwardPass();
y_hat = outputLayer->a; _output_layer->forwardPass();
_y_hat = _output_layer->a;
}
void MLPPMANN::_bind_methods() {
/*
ClassDB::bind_method(D_METHOD("get_input_set"), &MLPPMANN::get_input_set);
ClassDB::bind_method(D_METHOD("set_input_set", "val"), &MLPPMANN::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"), &MLPPMANN::get_output_set);
ClassDB::bind_method(D_METHOD("set_output_set", "val"), &MLPPMANN::set_output_set);
ADD_PROPERTY(PropertyInfo(Variant::OBJECT, "output_set", PROPERTY_HINT_RESOURCE_TYPE, "MLPPMatrix"), "set_output_set", "get_output_set");
*/
} }

72
mlpp/mann/mann.h
View File

@ -10,6 +10,13 @@
#include "core/math/math_defs.h" #include "core/math/math_defs.h"
#include "core/object/reference.h"
#include "../regularization/reg.h"
#include "../lin_alg/mlpp_matrix.h"
#include "../lin_alg/mlpp_vector.h"
#include "../hidden_layer/hidden_layer.h" #include "../hidden_layer/hidden_layer.h"
#include "../multi_output_layer/multi_output_layer.h" #include "../multi_output_layer/multi_output_layer.h"
@ -19,33 +26,56 @@
#include <string> #include <string>
#include <vector> #include <vector>
class MLPPMANN { class MLPPMANN : public Reference {
public: GDCLASS(MLPPMANN, Reference);
MLPPMANN(std::vector<std::vector<real_t>> inputSet, std::vector<std::vector<real_t>> outputSet);
~MLPPMANN();
std::vector<std::vector<real_t>> modelSetTest(std::vector<std::vector<real_t>> X);
std::vector<real_t> modelTest(std::vector<real_t> x);
void gradientDescent(real_t learning_rate, int max_epoch, bool UI = false);
real_t score();
void save(std::string fileName);
void addLayer(int n_hidden, std::string activation, std::string weightInit = "Default", std::string reg = "None", real_t lambda = 0.5, real_t alpha = 0.5); public:
void addOutputLayer(std::string activation, std::string loss, std::string weightInit = "Default", std::string reg = "None", real_t lambda = 0.5, real_t alpha = 0.5); /*
Ref<MLPPMatrix> get_input_set();
void set_input_set(const Ref<MLPPMatrix> &val);
Ref<MLPPMatrix> get_output_set();
void set_output_set(const Ref<MLPPMatrix> &val);
*/
std::vector<std::vector<real_t>> model_set_test(std::vector<std::vector<real_t>> X);
std::vector<real_t> model_test(std::vector<real_t> x);
void gradient_descent(real_t learning_rate, int max_epoch, bool ui = false);
real_t score();
void save(std::string file_name);
void add_layer(int n_hidden, std::string activation, std::string weight_init = "Default", std::string reg = "None", real_t lambda = 0.5, real_t alpha = 0.5);
void add_output_layer(std::string activation, std::string loss, std::string weight_init = "Default", std::string reg = "None", real_t lambda = 0.5, real_t alpha = 0.5);
bool is_initialized();
void initialize();
MLPPMANN(std::vector<std::vector<real_t>> p_input_set, std::vector<std::vector<real_t>> p_output_set);
MLPPMANN();
~MLPPMANN();
private: private:
real_t Cost(std::vector<std::vector<real_t>> y_hat, std::vector<std::vector<real_t>> y); real_t cost(std::vector<std::vector<real_t>> y_hat, std::vector<std::vector<real_t>> y);
void forwardPass();
std::vector<std::vector<real_t>> inputSet; void forward_pass();
std::vector<std::vector<real_t>> outputSet;
std::vector<std::vector<real_t>> y_hat;
std::vector<MLPPOldHiddenLayer> network; static void _bind_methods();
MLPPOldMultiOutputLayer *outputLayer;
int n; std::vector<std::vector<real_t>> _input_set;
int k; std::vector<std::vector<real_t>> _output_set;
int n_output; std::vector<std::vector<real_t>> _y_hat;
std::vector<MLPPOldHiddenLayer> _network;
MLPPOldMultiOutputLayer *_output_layer;
int _n;
int _k;
int _n_output;
bool _initialized;
}; };
#endif /* MANN_hpp */ #endif /* MANN_hpp */

2
mlpp/mlp/mlp.h
View File

@ -99,7 +99,7 @@ protected:
real_t lambda; /* Regularization Parameter */ real_t lambda; /* Regularization Parameter */
real_t alpha; /* This is the controlling param for Elastic Net*/ real_t alpha; /* This is the controlling param for Elastic Net*/
int _initialized; bool _initialized;
}; };
#endif /* MLP_hpp */ #endif /* MLP_hpp */

27
test/mlpp_tests.cpp
View File

@ -627,10 +627,16 @@ void MLPPTests::test_dynamically_sized_mann(bool ui) {
std::vector<std::vector<real_t>> inputSet = { { 1, 2, 3 }, { 2, 4, 6 }, { 3, 6, 9 }, { 4, 8, 12 } }; std::vector<std::vector<real_t>> inputSet = { { 1, 2, 3 }, { 2, 4, 6 }, { 3, 6, 9 }, { 4, 8, 12 } };
std::vector<std::vector<real_t>> outputSet = { { 1, 5 }, { 2, 10 }, { 3, 15 }, { 4, 20 } }; std::vector<std::vector<real_t>> outputSet = { { 1, 5 }, { 2, 10 }, { 3, 15 }, { 4, 20 } };
MLPPMANNOld mann_old(inputSet, outputSet);
mann_old.addOutputLayer("Linear", "MSE");
mann_old.gradientDescent(0.001, 80000, false);
alg.printMatrix(mann_old.modelSetTest(inputSet));
std::cout << "ACCURACY (old): " << 100 * mann_old.score() << "%" << std::endl;
MLPPMANN mann(inputSet, outputSet); MLPPMANN mann(inputSet, outputSet);
mann.addOutputLayer("Linear", "MSE"); mann.add_output_layer("Linear", "MSE");
mann.gradientDescent(0.001, 80000, 0); mann.gradient_descent(0.001, 80000, false);
alg.printMatrix(mann.modelSetTest(inputSet)); alg.printMatrix(mann.model_set_test(inputSet));
std::cout << "ACCURACY: " << 100 * mann.score() << "%" << std::endl; std::cout << "ACCURACY: " << 100 * mann.score() << "%" << std::endl;
} }
void MLPPTests::test_train_test_split_mann(bool ui) { void MLPPTests::test_train_test_split_mann(bool ui) {
@ -662,11 +668,18 @@ void MLPPTests::test_train_test_split_mann(bool ui) {
PLOG_MSG(split_data.test->get_input()->to_string()); PLOG_MSG(split_data.test->get_input()->to_string());
PLOG_MSG(split_data.test->get_output()->to_string()); PLOG_MSG(split_data.test->get_output()->to_string());
MLPPMANNOld mann_old(split_data.train->get_input()->to_std_vector(), split_data.train->get_output()->to_std_vector());
mann_old.addLayer(100, "RELU", "XavierNormal");
mann_old.addOutputLayer("Softmax", "CrossEntropy", "XavierNormal");
mann_old.gradientDescent(0.1, 80000, ui);
alg.printMatrix(mann_old.modelSetTest(split_data.test->get_input()->to_std_vector()));
std::cout << "ACCURACY (old): " << 100 * mann_old.score() << "%" << std::endl;
MLPPMANN mann(split_data.train->get_input()->to_std_vector(), split_data.train->get_output()->to_std_vector()); MLPPMANN mann(split_data.train->get_input()->to_std_vector(), split_data.train->get_output()->to_std_vector());
mann.addLayer(100, "RELU", "XavierNormal"); mann.add_layer(100, "RELU", "XavierNormal");
mann.addOutputLayer("Softmax", "CrossEntropy", "XavierNormal"); mann.add_output_layer("Softmax", "CrossEntropy", "XavierNormal");
mann.gradientDescent(0.1, 80000, ui); mann.gradient_descent(0.1, 80000, ui);
alg.printMatrix(mann.modelSetTest(split_data.test->get_input()->to_std_vector())); alg.printMatrix(mann.model_set_test(split_data.test->get_input()->to_std_vector()));
std::cout << "ACCURACY: " << 100 * mann.score() << "%" << std::endl; std::cout << "ACCURACY: " << 100 * mann.score() << "%" << std::endl;
} }