mirror of
https://github.com/Relintai/pmlpp.git
synced 2024-12-22 15:06:47 +01:00
MLPPProbitReg api rework.
This commit is contained in:
parent
27d197bf5d
commit
9d7fc44ca6
@ -18,8 +18,6 @@ Ref<MLPPMatrix> MLPPProbitReg::get_input_set() {
|
||||
}
|
||||
void MLPPProbitReg::set_input_set(const Ref<MLPPMatrix> &val) {
|
||||
_input_set = val;
|
||||
|
||||
_initialized = false;
|
||||
}
|
||||
|
||||
Ref<MLPPVector> MLPPProbitReg::get_output_set() {
|
||||
@ -27,8 +25,6 @@ Ref<MLPPVector> MLPPProbitReg::get_output_set() {
|
||||
}
|
||||
void MLPPProbitReg::set_output_set(const Ref<MLPPVector> &val) {
|
||||
_output_set = val;
|
||||
|
||||
_initialized = false;
|
||||
}
|
||||
|
||||
MLPPReg::RegularizationType MLPPProbitReg::get_reg() {
|
||||
@ -36,8 +32,6 @@ MLPPReg::RegularizationType MLPPProbitReg::get_reg() {
|
||||
}
|
||||
void MLPPProbitReg::set_reg(const MLPPReg::RegularizationType val) {
|
||||
_reg = val;
|
||||
|
||||
_initialized = false;
|
||||
}
|
||||
|
||||
real_t MLPPProbitReg::get_lambda() {
|
||||
@ -45,8 +39,6 @@ real_t MLPPProbitReg::get_lambda() {
|
||||
}
|
||||
void MLPPProbitReg::set_lambda(const real_t val) {
|
||||
_lambda = val;
|
||||
|
||||
_initialized = false;
|
||||
}
|
||||
|
||||
real_t MLPPProbitReg::get_alpha() {
|
||||
@ -54,25 +46,68 @@ real_t MLPPProbitReg::get_alpha() {
|
||||
}
|
||||
void MLPPProbitReg::set_alpha(const real_t val) {
|
||||
_alpha = val;
|
||||
}
|
||||
|
||||
_initialized = false;
|
||||
Ref<MLPPVector> MLPPProbitReg::data_z_get() const {
|
||||
return _z;
|
||||
}
|
||||
void MLPPProbitReg::data_z_set(const Ref<MLPPVector> &val) {
|
||||
if (!val.is_valid()) {
|
||||
return;
|
||||
}
|
||||
|
||||
_z = val;
|
||||
}
|
||||
|
||||
Ref<MLPPVector> MLPPProbitReg::data_y_hat_get() const {
|
||||
return _y_hat;
|
||||
}
|
||||
void MLPPProbitReg::data_y_hat_set(const Ref<MLPPVector> &val) {
|
||||
if (!val.is_valid()) {
|
||||
return;
|
||||
}
|
||||
|
||||
_y_hat = val;
|
||||
}
|
||||
|
||||
Ref<MLPPVector> MLPPProbitReg::data_weights_get() const {
|
||||
return _weights;
|
||||
}
|
||||
void MLPPProbitReg::data_weights_set(const Ref<MLPPVector> &val) {
|
||||
if (!val.is_valid()) {
|
||||
return;
|
||||
}
|
||||
|
||||
_weights = val;
|
||||
}
|
||||
|
||||
real_t MLPPProbitReg::data_bias_get() const {
|
||||
return _bias;
|
||||
}
|
||||
void MLPPProbitReg::data_bias_set(const real_t val) {
|
||||
_bias = val;
|
||||
}
|
||||
|
||||
Ref<MLPPVector> MLPPProbitReg::model_set_test(const Ref<MLPPMatrix> &X) {
|
||||
ERR_FAIL_COND_V(needs_init(), Ref<MLPPVector>());
|
||||
|
||||
return evaluatem(X);
|
||||
}
|
||||
|
||||
real_t MLPPProbitReg::model_test(const Ref<MLPPVector> &x) {
|
||||
ERR_FAIL_COND_V(needs_init(), 0);
|
||||
|
||||
return evaluatev(x);
|
||||
}
|
||||
|
||||
void MLPPProbitReg::gradient_descent(real_t learning_rate, int max_epoch, bool ui) {
|
||||
ERR_FAIL_COND(!_initialized);
|
||||
void MLPPProbitReg::train_gradient_descent(real_t learning_rate, int max_epoch, bool ui) {
|
||||
ERR_FAIL_COND(needs_init());
|
||||
|
||||
MLPPActivation avn;
|
||||
MLPPReg regularization;
|
||||
real_t cost_prev = 0;
|
||||
int epoch = 1;
|
||||
int n = _input_set->size().y;
|
||||
|
||||
forward_pass();
|
||||
|
||||
@ -82,11 +117,11 @@ void MLPPProbitReg::gradient_descent(real_t learning_rate, int max_epoch, bool u
|
||||
Ref<MLPPVector> error = _y_hat->subn(_output_set);
|
||||
|
||||
// Calculating the weight gradients
|
||||
_weights->sub(_input_set->transposen()->mult_vec(error->hadamard_productn(avn.gaussian_cdf_derivv(_z)))->scalar_multiplyn(learning_rate / _n));
|
||||
_weights->sub(_input_set->transposen()->mult_vec(error->hadamard_productn(avn.gaussian_cdf_derivv(_z)))->scalar_multiplyn(learning_rate / n));
|
||||
_weights = regularization.reg_weightsv(_weights, _lambda, _alpha, _reg);
|
||||
|
||||
// Calculating the bias gradients
|
||||
_bias -= learning_rate * error->hadamard_productn(avn.gaussian_cdf_derivv(_z))->sum_elements() / _n;
|
||||
_bias -= learning_rate * error->hadamard_productn(avn.gaussian_cdf_derivv(_z))->sum_elements() / n;
|
||||
|
||||
forward_pass();
|
||||
|
||||
@ -103,13 +138,14 @@ void MLPPProbitReg::gradient_descent(real_t learning_rate, int max_epoch, bool u
|
||||
}
|
||||
}
|
||||
|
||||
void MLPPProbitReg::mle(real_t learning_rate, int max_epoch, bool ui) {
|
||||
ERR_FAIL_COND(!_initialized);
|
||||
void MLPPProbitReg::train_mle(real_t learning_rate, int max_epoch, bool ui) {
|
||||
ERR_FAIL_COND(needs_init());
|
||||
|
||||
MLPPActivation avn;
|
||||
MLPPReg regularization;
|
||||
real_t cost_prev = 0;
|
||||
int epoch = 1;
|
||||
int n = _input_set->size().y;
|
||||
|
||||
forward_pass();
|
||||
|
||||
@ -119,11 +155,11 @@ void MLPPProbitReg::mle(real_t learning_rate, int max_epoch, bool ui) {
|
||||
Ref<MLPPVector> error = _output_set->subn(_y_hat);
|
||||
|
||||
// Calculating the weight gradients
|
||||
_weights->add(_input_set->transposen()->mult_vec(error->hadamard_productn(avn.gaussian_cdf_derivv(_z)))->scalar_multiplyn(learning_rate / _n));
|
||||
_weights->add(_input_set->transposen()->mult_vec(error->hadamard_productn(avn.gaussian_cdf_derivv(_z)))->scalar_multiplyn(learning_rate / n));
|
||||
_weights = regularization.reg_weightsv(_weights, _lambda, _alpha, _reg);
|
||||
|
||||
// Calculating the bias gradients
|
||||
_bias += learning_rate * error->hadamard_productn(avn.gaussian_cdf_derivv(_z))->sum_elements() / _n;
|
||||
_bias += learning_rate * error->hadamard_productn(avn.gaussian_cdf_derivv(_z))->sum_elements() / n;
|
||||
|
||||
forward_pass();
|
||||
|
||||
@ -140,14 +176,15 @@ void MLPPProbitReg::mle(real_t learning_rate, int max_epoch, bool ui) {
|
||||
}
|
||||
}
|
||||
|
||||
void MLPPProbitReg::sgd(real_t learning_rate, int max_epoch, bool ui) {
|
||||
ERR_FAIL_COND(!_initialized);
|
||||
void MLPPProbitReg::train_sgd(real_t learning_rate, int max_epoch, bool ui) {
|
||||
ERR_FAIL_COND(needs_init());
|
||||
|
||||
// NOTE: ∂y_hat/∂z is sparse
|
||||
MLPPActivation avn;
|
||||
MLPPReg regularization;
|
||||
real_t cost_prev = 0;
|
||||
int epoch = 1;
|
||||
int n = _input_set->size().y;
|
||||
|
||||
Ref<MLPPVector> input_set_row_tmp;
|
||||
input_set_row_tmp.instance();
|
||||
@ -163,7 +200,7 @@ void MLPPProbitReg::sgd(real_t learning_rate, int max_epoch, bool ui) {
|
||||
|
||||
std::random_device 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 output_index = distribution(generator);
|
||||
@ -205,20 +242,21 @@ void MLPPProbitReg::sgd(real_t learning_rate, int max_epoch, bool ui) {
|
||||
forward_pass();
|
||||
}
|
||||
|
||||
void MLPPProbitReg::mbgd(real_t learning_rate, int max_epoch, int mini_batch_size, bool ui) {
|
||||
ERR_FAIL_COND(!_initialized);
|
||||
void MLPPProbitReg::train_mbgd(real_t learning_rate, int max_epoch, int mini_batch_size, bool ui) {
|
||||
ERR_FAIL_COND(needs_init());
|
||||
|
||||
MLPPActivation avn;
|
||||
MLPPReg regularization;
|
||||
real_t cost_prev = 0;
|
||||
int epoch = 1;
|
||||
int n = _input_set->size().y;
|
||||
|
||||
Ref<MLPPVector> z_tmp;
|
||||
z_tmp.instance();
|
||||
z_tmp->resize(1);
|
||||
|
||||
// Creating the mini-batches
|
||||
int n_mini_batch = _n / mini_batch_size;
|
||||
int n_mini_batch = n / mini_batch_size;
|
||||
|
||||
MLPPUtilities::CreateMiniBatchMVBatch batches = MLPPUtilities::create_mini_batchesmv(_input_set, _output_set, n_mini_batch);
|
||||
|
||||
@ -262,89 +300,77 @@ void MLPPProbitReg::mbgd(real_t learning_rate, int max_epoch, int mini_batch_siz
|
||||
}
|
||||
|
||||
real_t MLPPProbitReg::score() {
|
||||
ERR_FAIL_COND_V(needs_init(), 0);
|
||||
|
||||
MLPPUtilities util;
|
||||
|
||||
return util.performance_vec(_y_hat, _output_set);
|
||||
}
|
||||
|
||||
void MLPPProbitReg::save(const String &file_name) {
|
||||
MLPPUtilities util;
|
||||
|
||||
//util.saveParameters(file_name, _weights, _bias);
|
||||
bool MLPPProbitReg::needs_init() const {
|
||||
if (!_input_set.is_valid()) {
|
||||
return true;
|
||||
}
|
||||
|
||||
bool MLPPProbitReg::is_initialized() {
|
||||
return _initialized;
|
||||
if (!_output_set.is_valid()) {
|
||||
return true;
|
||||
}
|
||||
|
||||
int n = _input_set->size().y;
|
||||
int k = _input_set->size().x;
|
||||
|
||||
if (_y_hat->size() != n) {
|
||||
return true;
|
||||
}
|
||||
|
||||
if (_weights->size() != k) {
|
||||
return true;
|
||||
}
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
void MLPPProbitReg::initialize() {
|
||||
if (_initialized) {
|
||||
return;
|
||||
}
|
||||
|
||||
ERR_FAIL_COND(!_input_set.is_valid() || !_output_set.is_valid());
|
||||
|
||||
_n = _input_set->size().y;
|
||||
_k = _input_set->size().x;
|
||||
int n = _input_set->size().y;
|
||||
int k = _input_set->size().x;
|
||||
|
||||
if (!_y_hat.is_valid()) {
|
||||
_y_hat.instance();
|
||||
}
|
||||
|
||||
_y_hat->resize(_n);
|
||||
_y_hat->resize(n);
|
||||
|
||||
MLPPUtilities util;
|
||||
|
||||
if (!_weights.is_valid()) {
|
||||
_weights.instance();
|
||||
}
|
||||
|
||||
_weights->resize(_k);
|
||||
_weights->resize(k);
|
||||
|
||||
util.weight_initializationv(_weights);
|
||||
_bias = util.bias_initializationr();
|
||||
|
||||
_initialized = true;
|
||||
}
|
||||
|
||||
MLPPProbitReg::MLPPProbitReg(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;
|
||||
_output_set = p_output_set;
|
||||
|
||||
_n = _input_set->size().y;
|
||||
_k = _input_set->size().x;
|
||||
|
||||
_reg = p_reg;
|
||||
_lambda = p_lambda;
|
||||
_alpha = p_alpha;
|
||||
|
||||
_z.instance();
|
||||
_y_hat.instance();
|
||||
_y_hat->resize(_n);
|
||||
|
||||
MLPPUtilities util;
|
||||
|
||||
_weights.instance();
|
||||
_weights->resize(_k);
|
||||
_bias = 0;
|
||||
|
||||
util.weight_initializationv(_weights);
|
||||
_bias = util.bias_initializationr();
|
||||
|
||||
_initialized = true;
|
||||
initialize();
|
||||
}
|
||||
|
||||
MLPPProbitReg::MLPPProbitReg() {
|
||||
_y_hat.instance();
|
||||
|
||||
_bias = 0;
|
||||
|
||||
_n = 0;
|
||||
_k = 0;
|
||||
|
||||
// Regularization Params
|
||||
_reg = MLPPReg::REGULARIZATION_TYPE_NONE;
|
||||
_lambda = 0.5;
|
||||
_alpha = 0.5;
|
||||
|
||||
_initialized = false;
|
||||
_z.instance();
|
||||
_y_hat.instance();
|
||||
_weights.instance();
|
||||
_bias = 0;
|
||||
}
|
||||
MLPPProbitReg::~MLPPProbitReg() {
|
||||
}
|
||||
@ -405,18 +431,33 @@ void MLPPProbitReg::_bind_methods() {
|
||||
ClassDB::bind_method(D_METHOD("set_alpha", "val"), &MLPPProbitReg::set_alpha);
|
||||
ADD_PROPERTY(PropertyInfo(Variant::REAL, "alpha"), "set_alpha", "get_alpha");
|
||||
|
||||
ADD_GROUP("Data", "data");
|
||||
ClassDB::bind_method(D_METHOD("data_z_get"), &MLPPProbitReg::data_z_get);
|
||||
ClassDB::bind_method(D_METHOD("data_z_set", "val"), &MLPPProbitReg::set_output_set);
|
||||
ADD_PROPERTY(PropertyInfo(Variant::OBJECT, "data_z", PROPERTY_HINT_RESOURCE_TYPE, "MLPPVector"), "data_z_set", "data_z_get");
|
||||
|
||||
ClassDB::bind_method(D_METHOD("data_y_hat_get"), &MLPPProbitReg::data_y_hat_get);
|
||||
ClassDB::bind_method(D_METHOD("data_y_hat_set", "val"), &MLPPProbitReg::data_y_hat_set);
|
||||
ADD_PROPERTY(PropertyInfo(Variant::OBJECT, "data_y_hat", PROPERTY_HINT_RESOURCE_TYPE, "MLPPVector"), "data_y_hat_set", "data_y_hat_get");
|
||||
|
||||
ClassDB::bind_method(D_METHOD("data_weights_get"), &MLPPProbitReg::data_weights_get);
|
||||
ClassDB::bind_method(D_METHOD("data_weights_set", "val"), &MLPPProbitReg::data_weights_set);
|
||||
ADD_PROPERTY(PropertyInfo(Variant::OBJECT, "data_weights", PROPERTY_HINT_RESOURCE_TYPE, "MLPPVector"), "data_weights_set", "data_weights_get");
|
||||
|
||||
ClassDB::bind_method(D_METHOD("data_bias_get"), &MLPPProbitReg::data_bias_get);
|
||||
ClassDB::bind_method(D_METHOD("data_bias_set", "val"), &MLPPProbitReg::data_bias_set);
|
||||
ADD_PROPERTY(PropertyInfo(Variant::REAL, "data_bias"), "data_bias_set", "data_bias_get");
|
||||
|
||||
ClassDB::bind_method(D_METHOD("model_set_test", "X"), &MLPPProbitReg::model_set_test);
|
||||
ClassDB::bind_method(D_METHOD("model_test", "x"), &MLPPProbitReg::model_test);
|
||||
|
||||
ClassDB::bind_method(D_METHOD("gradient_descent", "learning_rate", "max_epoch", "ui"), &MLPPProbitReg::gradient_descent, 0, false);
|
||||
ClassDB::bind_method(D_METHOD("mle", "learning_rate", "max_epoch", "ui"), &MLPPProbitReg::mle, 0, false);
|
||||
ClassDB::bind_method(D_METHOD("sgd", "learning_rate", "max_epoch", "ui"), &MLPPProbitReg::sgd, 0, false);
|
||||
ClassDB::bind_method(D_METHOD("mbgd", "learning_rate", "max_epoch", "mini_batch_size", "ui"), &MLPPProbitReg::mbgd, false);
|
||||
ClassDB::bind_method(D_METHOD("train_gradient_descent", "learning_rate", "max_epoch", "ui"), &MLPPProbitReg::train_gradient_descent, 0, false);
|
||||
ClassDB::bind_method(D_METHOD("train_mle", "learning_rate", "max_epoch", "ui"), &MLPPProbitReg::train_mle, 0, false);
|
||||
ClassDB::bind_method(D_METHOD("train_sgd", "learning_rate", "max_epoch", "ui"), &MLPPProbitReg::train_sgd, 0, false);
|
||||
ClassDB::bind_method(D_METHOD("train_mbgd", "learning_rate", "max_epoch", "mini_batch_size", "ui"), &MLPPProbitReg::train_mbgd, false);
|
||||
|
||||
ClassDB::bind_method(D_METHOD("score"), &MLPPProbitReg::score);
|
||||
|
||||
ClassDB::bind_method(D_METHOD("save", "file_name"), &MLPPProbitReg::save);
|
||||
|
||||
ClassDB::bind_method(D_METHOD("is_initialized"), &MLPPProbitReg::is_initialized);
|
||||
ClassDB::bind_method(D_METHOD("needs_init"), &MLPPProbitReg::needs_init);
|
||||
ClassDB::bind_method(D_METHOD("initialize"), &MLPPProbitReg::initialize);
|
||||
}
|
||||
|
||||
@ -10,15 +10,15 @@
|
||||
|
||||
#include "core/math/math_defs.h"
|
||||
|
||||
#include "core/object/reference.h"
|
||||
#include "core/object/resource.h"
|
||||
|
||||
#include "../lin_alg/mlpp_matrix.h"
|
||||
#include "../lin_alg/mlpp_vector.h"
|
||||
|
||||
#include "../regularization/reg.h"
|
||||
|
||||
class MLPPProbitReg : public Reference {
|
||||
GDCLASS(MLPPProbitReg, Reference);
|
||||
class MLPPProbitReg : public Resource {
|
||||
GDCLASS(MLPPProbitReg, Resource);
|
||||
|
||||
public:
|
||||
Ref<MLPPMatrix> get_input_set();
|
||||
@ -36,19 +36,29 @@ public:
|
||||
real_t get_alpha();
|
||||
void set_alpha(const real_t val);
|
||||
|
||||
Ref<MLPPVector> data_z_get() const;
|
||||
void data_z_set(const Ref<MLPPVector> &val);
|
||||
|
||||
Ref<MLPPVector> data_y_hat_get() const;
|
||||
void data_y_hat_set(const Ref<MLPPVector> &val);
|
||||
|
||||
Ref<MLPPVector> data_weights_get() const;
|
||||
void data_weights_set(const Ref<MLPPVector> &val);
|
||||
|
||||
real_t data_bias_get() const;
|
||||
void data_bias_set(const real_t val);
|
||||
|
||||
Ref<MLPPVector> model_set_test(const Ref<MLPPMatrix> &X);
|
||||
real_t model_test(const Ref<MLPPVector> &x);
|
||||
|
||||
void gradient_descent(real_t learning_rate, int max_epoch = 0, bool ui = false);
|
||||
void mle(real_t learning_rate, int max_epoch = 0, bool ui = false);
|
||||
void sgd(real_t learning_rate, int max_epoch = 0, bool ui = false);
|
||||
void mbgd(real_t learning_rate, int max_epoch, int mini_batch_size, bool ui = false);
|
||||
void train_gradient_descent(real_t learning_rate, int max_epoch = 0, bool ui = false);
|
||||
void train_mle(real_t learning_rate, int max_epoch = 0, bool ui = false);
|
||||
void train_sgd(real_t learning_rate, int max_epoch = 0, bool ui = false);
|
||||
void train_mbgd(real_t learning_rate, int max_epoch, int mini_batch_size, bool ui = false);
|
||||
|
||||
real_t score();
|
||||
|
||||
void save(const String &file_name);
|
||||
|
||||
bool is_initialized();
|
||||
bool needs_init() const;
|
||||
void initialize();
|
||||
|
||||
MLPPProbitReg(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);
|
||||
@ -80,11 +90,6 @@ protected:
|
||||
Ref<MLPPVector> _y_hat;
|
||||
Ref<MLPPVector> _weights;
|
||||
real_t _bias;
|
||||
|
||||
int _n;
|
||||
int _k;
|
||||
|
||||
bool _initialized;
|
||||
};
|
||||
|
||||
#endif /* ProbitReg_hpp */
|
||||
|
||||
@ -283,7 +283,7 @@ void MLPPTests::test_probit_regression(bool ui) {
|
||||
Ref<MLPPDataSimple> dt = data.load_breast_cancer(_breast_cancer_data_path);
|
||||
|
||||
MLPPProbitReg model(dt->get_input(), dt->get_output());
|
||||
model.sgd(0.001, 10000, ui);
|
||||
model.train_sgd(0.001, 10000, ui);
|
||||
PLOG_MSG(model.model_set_test(dt->get_input())->to_string());
|
||||
PLOG_MSG("ACCURACY: " + String::num(100 * model.score()) + "%");
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user