mirror of
https://github.com/Relintai/pmlpp.git
synced 2024-12-22 15:06:47 +01:00
863 lines
28 KiB
C++
863 lines
28 KiB
C++
//
|
|
// Reg.cpp
|
|
//
|
|
// Created by Marc Melikyan on 1/16/21.
|
|
//
|
|
|
|
#include "utilities.h"
|
|
|
|
#include "core/log/logger.h"
|
|
#include "core/math/math_funcs.h"
|
|
#include "core/math/random_pcg.h"
|
|
|
|
#include <fstream>
|
|
#include <iostream>
|
|
#include <random>
|
|
#include <string>
|
|
|
|
std::vector<real_t> MLPPUtilities::weightInitialization(int n, std::string type) {
|
|
std::random_device rd;
|
|
std::default_random_engine generator(rd());
|
|
|
|
std::vector<real_t> weights;
|
|
for (int i = 0; i < n; i++) {
|
|
if (type == "XavierNormal") {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(2 / (n + 1)));
|
|
weights.push_back(distribution(generator));
|
|
} else if (type == "XavierUniform") {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(6 / (n + 1)), sqrt(6 / (n + 1)));
|
|
weights.push_back(distribution(generator));
|
|
} else if (type == "HeNormal") {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(2 / n));
|
|
weights.push_back(distribution(generator));
|
|
} else if (type == "HeUniform") {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(6 / n), sqrt(6 / n));
|
|
weights.push_back(distribution(generator));
|
|
} else if (type == "LeCunNormal") {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(1 / n));
|
|
weights.push_back(distribution(generator));
|
|
} else if (type == "LeCunUniform") {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(3 / n), sqrt(3 / n));
|
|
weights.push_back(distribution(generator));
|
|
} else if (type == "Uniform") {
|
|
std::uniform_real_distribution<real_t> distribution(-1 / sqrt(n), 1 / sqrt(n));
|
|
weights.push_back(distribution(generator));
|
|
} else {
|
|
std::uniform_real_distribution<real_t> distribution(0, 1);
|
|
weights.push_back(distribution(generator));
|
|
}
|
|
}
|
|
return weights;
|
|
}
|
|
|
|
real_t MLPPUtilities::biasInitialization() {
|
|
std::random_device rd;
|
|
std::default_random_engine generator(rd());
|
|
std::uniform_real_distribution<real_t> distribution(0, 1);
|
|
|
|
return distribution(generator);
|
|
}
|
|
|
|
std::vector<std::vector<real_t>> MLPPUtilities::weightInitialization(int n, int m, std::string type) {
|
|
std::random_device rd;
|
|
std::default_random_engine generator(rd());
|
|
|
|
std::vector<std::vector<real_t>> weights;
|
|
weights.resize(n);
|
|
|
|
for (int i = 0; i < n; i++) {
|
|
for (int j = 0; j < m; j++) {
|
|
if (type == "XavierNormal") {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(2 / (n + m)));
|
|
weights[i].push_back(distribution(generator));
|
|
} else if (type == "XavierUniform") {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(6 / (n + m)), sqrt(6 / (n + m)));
|
|
weights[i].push_back(distribution(generator));
|
|
} else if (type == "HeNormal") {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(2 / n));
|
|
weights[i].push_back(distribution(generator));
|
|
} else if (type == "HeUniform") {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(6 / n), sqrt(6 / n));
|
|
weights[i].push_back(distribution(generator));
|
|
} else if (type == "LeCunNormal") {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(1 / n));
|
|
weights[i].push_back(distribution(generator));
|
|
} else if (type == "LeCunUniform") {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(3 / n), sqrt(3 / n));
|
|
weights[i].push_back(distribution(generator));
|
|
} else if (type == "Uniform") {
|
|
std::uniform_real_distribution<real_t> distribution(-1 / sqrt(n), 1 / sqrt(n));
|
|
weights[i].push_back(distribution(generator));
|
|
} else {
|
|
std::uniform_real_distribution<real_t> distribution(0, 1);
|
|
weights[i].push_back(distribution(generator));
|
|
}
|
|
}
|
|
}
|
|
return weights;
|
|
}
|
|
|
|
std::vector<real_t> MLPPUtilities::biasInitialization(int n) {
|
|
std::vector<real_t> bias;
|
|
std::random_device rd;
|
|
std::default_random_engine generator(rd());
|
|
std::uniform_real_distribution<real_t> distribution(0, 1);
|
|
|
|
for (int i = 0; i < n; i++) {
|
|
bias.push_back(distribution(generator));
|
|
}
|
|
return bias;
|
|
}
|
|
|
|
void MLPPUtilities::weight_initializationv(Ref<MLPPVector> weights, WeightDistributionType type) {
|
|
ERR_FAIL_COND(!weights.is_valid());
|
|
|
|
int n = weights->size();
|
|
real_t *weights_ptr = weights->ptrw();
|
|
|
|
RandomPCG rnd;
|
|
rnd.randomize();
|
|
|
|
std::random_device rd;
|
|
std::default_random_engine generator(rd());
|
|
|
|
switch (type) {
|
|
case WEIGHT_DISTRIBUTION_TYPE_DEFAULT: {
|
|
std::uniform_real_distribution<real_t> distribution(0, 1);
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_XAVIER_NORMAL: {
|
|
std::normal_distribution<real_t> distribution(0, Math::sqrt(2.0 / (n + 1.0)));
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_XAVIER_UNIFORM: {
|
|
std::uniform_real_distribution<real_t> distribution(-Math::sqrt(6.0 / (n + 1.0)), Math::sqrt(6.0 / (n + 1.0)));
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_HE_NORMAL: {
|
|
std::normal_distribution<real_t> distribution(0, Math::sqrt(2.0 / n));
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_HE_UNIFORM: {
|
|
std::uniform_real_distribution<real_t> distribution(-Math::sqrt(6.0 / n), Math::sqrt(6.0 / n));
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_LE_CUN_NORMAL: {
|
|
std::normal_distribution<real_t> distribution(0, Math::sqrt(1.0 / n));
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_LE_CUN_UNIFORM: {
|
|
std::uniform_real_distribution<real_t> distribution(-Math::sqrt(3.0 / n), Math::sqrt(3.0 / n));
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_UNIFORM: {
|
|
std::uniform_real_distribution<real_t> distribution(-1.0 / Math::sqrt(static_cast<real_t>(n)), 1.0 / Math::sqrt(static_cast<real_t>(n)));
|
|
|
|
for (int i = 0; i < n; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
void MLPPUtilities::weight_initializationm(Ref<MLPPMatrix> weights, WeightDistributionType type) {
|
|
ERR_FAIL_COND(!weights.is_valid());
|
|
|
|
int n = weights->size().x;
|
|
int m = weights->size().y;
|
|
int data_size = weights->data_size();
|
|
real_t *weights_ptr = weights->ptrw();
|
|
|
|
RandomPCG rnd;
|
|
rnd.randomize();
|
|
|
|
std::random_device rd;
|
|
std::default_random_engine generator(rd());
|
|
|
|
switch (type) {
|
|
case WEIGHT_DISTRIBUTION_TYPE_DEFAULT: {
|
|
std::uniform_real_distribution<real_t> distribution(0, 1);
|
|
|
|
for (int i = 0; i < data_size; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_XAVIER_NORMAL: {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(2 / (n + m)));
|
|
|
|
for (int i = 0; i < data_size; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_XAVIER_UNIFORM: {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(6 / (n + m)), sqrt(6 / (n + m)));
|
|
|
|
for (int i = 0; i < data_size; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_HE_NORMAL: {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(2 / n));
|
|
|
|
for (int i = 0; i < data_size; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_HE_UNIFORM: {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(6 / n), sqrt(6 / n));
|
|
|
|
for (int i = 0; i < data_size; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_LE_CUN_NORMAL: {
|
|
std::normal_distribution<real_t> distribution(0, sqrt(1 / n));
|
|
|
|
for (int i = 0; i < data_size; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_LE_CUN_UNIFORM: {
|
|
std::uniform_real_distribution<real_t> distribution(-sqrt(3 / n), sqrt(3 / n));
|
|
|
|
for (int i = 0; i < data_size; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
case WEIGHT_DISTRIBUTION_TYPE_UNIFORM: {
|
|
std::uniform_real_distribution<real_t> distribution(-1 / sqrt(n), 1 / sqrt(n));
|
|
|
|
for (int i = 0; i < data_size; ++i) {
|
|
weights_ptr[i] = distribution(generator);
|
|
}
|
|
} break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
real_t MLPPUtilities::bias_initializationr() {
|
|
std::random_device rd;
|
|
std::default_random_engine generator(rd());
|
|
std::uniform_real_distribution<real_t> distribution(0, 1);
|
|
|
|
return distribution(generator);
|
|
}
|
|
void MLPPUtilities::bias_initializationv(Ref<MLPPVector> z) {
|
|
ERR_FAIL_COND(!z.is_valid());
|
|
|
|
std::vector<real_t> bias;
|
|
std::random_device rd;
|
|
std::default_random_engine generator(rd());
|
|
std::uniform_real_distribution<real_t> distribution(0, 1);
|
|
|
|
int n = z->size();
|
|
|
|
for (int i = 0; i < n; i++) {
|
|
bias.push_back(distribution(generator));
|
|
}
|
|
}
|
|
|
|
real_t MLPPUtilities::performance(std::vector<real_t> y_hat, std::vector<real_t> outputSet) {
|
|
real_t correct = 0;
|
|
for (uint32_t i = 0; i < y_hat.size(); i++) {
|
|
if (std::round(y_hat[i]) == outputSet[i]) {
|
|
correct++;
|
|
}
|
|
}
|
|
return correct / y_hat.size();
|
|
}
|
|
|
|
real_t MLPPUtilities::performance(std::vector<std::vector<real_t>> y_hat, std::vector<std::vector<real_t>> y) {
|
|
real_t correct = 0;
|
|
for (uint32_t i = 0; i < y_hat.size(); i++) {
|
|
uint32_t sub_correct = 0;
|
|
for (uint32_t j = 0; j < y_hat[i].size(); j++) {
|
|
if (std::round(y_hat[i][j]) == y[i][j]) {
|
|
sub_correct++;
|
|
}
|
|
if (sub_correct == y_hat[0].size()) {
|
|
correct++;
|
|
}
|
|
}
|
|
}
|
|
return correct / y_hat.size();
|
|
}
|
|
|
|
real_t MLPPUtilities::performance_vec(const Ref<MLPPVector> &y_hat, const Ref<MLPPVector> &output_set) {
|
|
ERR_FAIL_COND_V(!y_hat.is_valid(), 0);
|
|
ERR_FAIL_COND_V(!output_set.is_valid(), 0);
|
|
|
|
real_t correct = 0;
|
|
for (int i = 0; i < y_hat->size(); i++) {
|
|
if (Math::is_equal_approx(y_hat->get_element(i), output_set->get_element(i))) {
|
|
correct++;
|
|
}
|
|
}
|
|
return correct / y_hat->size();
|
|
}
|
|
real_t MLPPUtilities::performance_mat(const Ref<MLPPMatrix> &y_hat, const Ref<MLPPMatrix> &y) {
|
|
ERR_FAIL_COND_V(!y_hat.is_valid(), 0);
|
|
ERR_FAIL_COND_V(!y.is_valid(), 0);
|
|
|
|
real_t correct = 0;
|
|
for (int i = 0; i < y_hat->size().y; i++) {
|
|
int sub_correct = 0;
|
|
|
|
for (int j = 0; j < y_hat->size().x; j++) {
|
|
if (Math::round(y_hat->get_element(i, j)) == y->get_element(i, j)) {
|
|
sub_correct++;
|
|
}
|
|
|
|
if (sub_correct == y_hat->size().x) {
|
|
correct++;
|
|
}
|
|
}
|
|
}
|
|
return correct / y_hat->size().y;
|
|
}
|
|
real_t MLPPUtilities::performance_pool_int_array_vec(PoolIntArray y_hat, const Ref<MLPPVector> &output_set) {
|
|
ERR_FAIL_COND_V(!output_set.is_valid(), 0);
|
|
|
|
real_t correct = 0;
|
|
for (int i = 0; i < y_hat.size(); i++) {
|
|
if (y_hat[i] == Math::round(output_set->get_element(i))) {
|
|
correct++;
|
|
}
|
|
}
|
|
return correct / y_hat.size();
|
|
}
|
|
|
|
void MLPPUtilities::saveParameters(std::string fileName, std::vector<real_t> weights, real_t bias, bool app, int layer) {
|
|
std::string layer_info = "";
|
|
std::ofstream saveFile;
|
|
|
|
if (layer > -1) {
|
|
layer_info = " for layer " + std::to_string(layer);
|
|
}
|
|
|
|
if (app) {
|
|
saveFile.open(fileName.c_str(), std::ios_base::app);
|
|
} else {
|
|
saveFile.open(fileName.c_str());
|
|
}
|
|
|
|
if (!saveFile.is_open()) {
|
|
std::cout << fileName << " failed to open." << std::endl;
|
|
}
|
|
|
|
saveFile << "Weight(s)" << layer_info << std::endl;
|
|
for (uint32_t i = 0; i < weights.size(); i++) {
|
|
saveFile << weights[i] << std::endl;
|
|
}
|
|
saveFile << "Bias" << layer_info << std::endl;
|
|
saveFile << bias << std::endl;
|
|
|
|
saveFile.close();
|
|
}
|
|
|
|
void MLPPUtilities::saveParameters(std::string fileName, std::vector<real_t> weights, std::vector<real_t> initial, real_t bias, bool app, int layer) {
|
|
std::string layer_info = "";
|
|
std::ofstream saveFile;
|
|
|
|
if (layer > -1) {
|
|
layer_info = " for layer " + std::to_string(layer);
|
|
}
|
|
|
|
if (app) {
|
|
saveFile.open(fileName.c_str(), std::ios_base::app);
|
|
} else {
|
|
saveFile.open(fileName.c_str());
|
|
}
|
|
|
|
if (!saveFile.is_open()) {
|
|
std::cout << fileName << " failed to open." << std::endl;
|
|
}
|
|
|
|
saveFile << "Weight(s)" << layer_info << std::endl;
|
|
for (uint32_t i = 0; i < weights.size(); i++) {
|
|
saveFile << weights[i] << std::endl;
|
|
}
|
|
|
|
saveFile << "Initial(s)" << layer_info << std::endl;
|
|
for (uint32_t i = 0; i < initial.size(); i++) {
|
|
saveFile << initial[i] << std::endl;
|
|
}
|
|
|
|
saveFile << "Bias" << layer_info << std::endl;
|
|
saveFile << bias << std::endl;
|
|
|
|
saveFile.close();
|
|
}
|
|
|
|
void MLPPUtilities::saveParameters(std::string fileName, std::vector<std::vector<real_t>> weights, std::vector<real_t> bias, bool app, int layer) {
|
|
std::string layer_info = "";
|
|
std::ofstream saveFile;
|
|
|
|
if (layer > -1) {
|
|
layer_info = " for layer " + std::to_string(layer);
|
|
}
|
|
|
|
if (app) {
|
|
saveFile.open(fileName.c_str(), std::ios_base::app);
|
|
} else {
|
|
saveFile.open(fileName.c_str());
|
|
}
|
|
|
|
if (!saveFile.is_open()) {
|
|
std::cout << fileName << " failed to open." << std::endl;
|
|
}
|
|
|
|
saveFile << "Weight(s)" << layer_info << std::endl;
|
|
for (uint32_t i = 0; i < weights.size(); i++) {
|
|
for (uint32_t j = 0; j < weights[i].size(); j++) {
|
|
saveFile << weights[i][j] << std::endl;
|
|
}
|
|
}
|
|
saveFile << "Bias(es)" << layer_info << std::endl;
|
|
for (uint32_t i = 0; i < bias.size(); i++) {
|
|
saveFile << bias[i] << std::endl;
|
|
}
|
|
|
|
saveFile.close();
|
|
}
|
|
|
|
void MLPPUtilities::UI(std::vector<real_t> weights, real_t bias) {
|
|
std::cout << "Values of the weight(s):" << std::endl;
|
|
for (uint32_t i = 0; i < weights.size(); i++) {
|
|
std::cout << weights[i] << std::endl;
|
|
}
|
|
std::cout << "Value of the bias:" << std::endl;
|
|
std::cout << bias << std::endl;
|
|
}
|
|
|
|
void MLPPUtilities::UI(std::vector<std::vector<real_t>> weights, std::vector<real_t> bias) {
|
|
std::cout << "Values of the weight(s):" << std::endl;
|
|
for (uint32_t i = 0; i < weights.size(); i++) {
|
|
for (uint32_t j = 0; j < weights[i].size(); j++) {
|
|
std::cout << weights[i][j] << std::endl;
|
|
}
|
|
}
|
|
std::cout << "Value of the biases:" << std::endl;
|
|
for (uint32_t i = 0; i < bias.size(); i++) {
|
|
std::cout << bias[i] << std::endl;
|
|
}
|
|
}
|
|
|
|
void MLPPUtilities::UI(std::vector<real_t> weights, std::vector<real_t> initial, real_t bias) {
|
|
std::cout << "Values of the weight(s):" << std::endl;
|
|
for (uint32_t i = 0; i < weights.size(); i++) {
|
|
std::cout << weights[i] << std::endl;
|
|
}
|
|
std::cout << "Values of the initial(s):" << std::endl;
|
|
for (uint32_t i = 0; i < initial.size(); i++) {
|
|
std::cout << initial[i] << std::endl;
|
|
}
|
|
std::cout << "Value of the bias:" << std::endl;
|
|
std::cout << bias << std::endl;
|
|
}
|
|
|
|
void MLPPUtilities::print_ui_vb(Ref<MLPPVector> weights, real_t bias) {
|
|
String str = "Values of the weight(s):\n";
|
|
str += weights->to_string();
|
|
str += "\nValue of the bias:\n";
|
|
str += String::num(bias);
|
|
|
|
PLOG_MSG(str);
|
|
}
|
|
void MLPPUtilities::print_ui_vib(Ref<MLPPVector> weights, Ref<MLPPVector> initial, real_t bias) {
|
|
String str = "Values of the weight(s):\n";
|
|
str += weights->to_string();
|
|
|
|
str += "\nValues of the initial(s):\n";
|
|
str += initial->to_string();
|
|
|
|
str += "\nValue of the bias:\n";
|
|
str += String::num(bias);
|
|
|
|
PLOG_MSG(str);
|
|
}
|
|
void MLPPUtilities::print_ui_mb(Ref<MLPPMatrix> weights, Ref<MLPPVector> bias) {
|
|
String str = "Values of the weight(s):\n";
|
|
str += weights->to_string();
|
|
|
|
str += "\nValue of the biased:\n";
|
|
str += bias->to_string();
|
|
|
|
PLOG_MSG(str);
|
|
}
|
|
|
|
void MLPPUtilities::CostInfo(int epoch, real_t cost_prev, real_t Cost) {
|
|
std::cout << "-----------------------------------" << std::endl;
|
|
std::cout << "This is epoch: " << epoch << std::endl;
|
|
std::cout << "The cost function has been minimized by " << cost_prev - Cost << std::endl;
|
|
std::cout << "Current Cost:" << std::endl;
|
|
std::cout << Cost << std::endl;
|
|
}
|
|
|
|
void MLPPUtilities::cost_info(int epoch, real_t cost_prev, real_t cost) {
|
|
String str = "This is epoch: " + itos(epoch) + ",";
|
|
str += "The cost function has been minimized by " + String::num(cost_prev - cost);
|
|
str += ", Current Cost:" + String::num(cost);
|
|
|
|
PLOG_MSG(str);
|
|
}
|
|
|
|
std::vector<std::vector<std::vector<real_t>>> MLPPUtilities::createMiniBatches(std::vector<std::vector<real_t>> inputSet, int n_mini_batch) {
|
|
int n = inputSet.size();
|
|
|
|
std::vector<std::vector<std::vector<real_t>>> inputMiniBatches;
|
|
|
|
// Creating the mini-batches
|
|
for (int i = 0; i < n_mini_batch; i++) {
|
|
std::vector<std::vector<real_t>> currentInputSet;
|
|
for (int j = 0; j < n / n_mini_batch; j++) {
|
|
currentInputSet.push_back(inputSet[n / n_mini_batch * i + j]);
|
|
}
|
|
inputMiniBatches.push_back(currentInputSet);
|
|
}
|
|
|
|
if (real_t(n) / real_t(n_mini_batch) - int(n / n_mini_batch) != 0) {
|
|
for (int i = 0; i < n - n / n_mini_batch * n_mini_batch; i++) {
|
|
inputMiniBatches[n_mini_batch - 1].push_back(inputSet[n / n_mini_batch * n_mini_batch + i]);
|
|
}
|
|
}
|
|
return inputMiniBatches;
|
|
}
|
|
|
|
std::tuple<std::vector<std::vector<std::vector<real_t>>>, std::vector<std::vector<real_t>>> MLPPUtilities::createMiniBatches(std::vector<std::vector<real_t>> inputSet, std::vector<real_t> outputSet, int n_mini_batch) {
|
|
int n = inputSet.size();
|
|
|
|
std::vector<std::vector<std::vector<real_t>>> inputMiniBatches;
|
|
std::vector<std::vector<real_t>> outputMiniBatches;
|
|
|
|
for (int i = 0; i < n_mini_batch; i++) {
|
|
std::vector<std::vector<real_t>> currentInputSet;
|
|
std::vector<real_t> currentOutputSet;
|
|
for (int j = 0; j < n / n_mini_batch; j++) {
|
|
currentInputSet.push_back(inputSet[n / n_mini_batch * i + j]);
|
|
currentOutputSet.push_back(outputSet[n / n_mini_batch * i + j]);
|
|
}
|
|
inputMiniBatches.push_back(currentInputSet);
|
|
outputMiniBatches.push_back(currentOutputSet);
|
|
}
|
|
|
|
if (real_t(n) / real_t(n_mini_batch) - int(n / n_mini_batch) != 0) {
|
|
for (int i = 0; i < n - n / n_mini_batch * n_mini_batch; i++) {
|
|
inputMiniBatches[n_mini_batch - 1].push_back(inputSet[n / n_mini_batch * n_mini_batch + i]);
|
|
outputMiniBatches[n_mini_batch - 1].push_back(outputSet[n / n_mini_batch * n_mini_batch + i]);
|
|
}
|
|
}
|
|
return { inputMiniBatches, outputMiniBatches };
|
|
}
|
|
|
|
std::tuple<std::vector<std::vector<std::vector<real_t>>>, std::vector<std::vector<std::vector<real_t>>>> MLPPUtilities::createMiniBatches(std::vector<std::vector<real_t>> inputSet, std::vector<std::vector<real_t>> outputSet, int n_mini_batch) {
|
|
int n = inputSet.size();
|
|
|
|
std::vector<std::vector<std::vector<real_t>>> inputMiniBatches;
|
|
std::vector<std::vector<std::vector<real_t>>> outputMiniBatches;
|
|
|
|
for (int i = 0; i < n_mini_batch; i++) {
|
|
std::vector<std::vector<real_t>> currentInputSet;
|
|
std::vector<std::vector<real_t>> currentOutputSet;
|
|
for (int j = 0; j < n / n_mini_batch; j++) {
|
|
currentInputSet.push_back(inputSet[n / n_mini_batch * i + j]);
|
|
currentOutputSet.push_back(outputSet[n / n_mini_batch * i + j]);
|
|
}
|
|
inputMiniBatches.push_back(currentInputSet);
|
|
outputMiniBatches.push_back(currentOutputSet);
|
|
}
|
|
|
|
if (real_t(n) / real_t(n_mini_batch) - int(n / n_mini_batch) != 0) {
|
|
for (int i = 0; i < n - n / n_mini_batch * n_mini_batch; i++) {
|
|
inputMiniBatches[n_mini_batch - 1].push_back(inputSet[n / n_mini_batch * n_mini_batch + i]);
|
|
outputMiniBatches[n_mini_batch - 1].push_back(outputSet[n / n_mini_batch * n_mini_batch + i]);
|
|
}
|
|
}
|
|
return { inputMiniBatches, outputMiniBatches };
|
|
}
|
|
|
|
Vector<Ref<MLPPMatrix>> MLPPUtilities::create_mini_batchesm(const Ref<MLPPMatrix> &input_set, int n_mini_batch) {
|
|
Size2i size = input_set->size();
|
|
|
|
int n = size.y;
|
|
int mini_batch_element_count = n / n_mini_batch;
|
|
|
|
Ref<MLPPVector> row_tmp;
|
|
row_tmp.instance();
|
|
row_tmp->resize(size.x);
|
|
|
|
Vector<Ref<MLPPMatrix>> input_mini_batches;
|
|
|
|
// Creating the mini-batches
|
|
for (int i = 0; i < n_mini_batch; i++) {
|
|
int mini_batch_start_offset = n_mini_batch * i;
|
|
Ref<MLPPMatrix> current_input_set;
|
|
current_input_set.instance();
|
|
current_input_set->resize(Size2i(size.x, mini_batch_element_count));
|
|
|
|
for (int j = 0; j < mini_batch_element_count; j++) {
|
|
input_set->get_row_into_mlpp_vector(mini_batch_start_offset + j, row_tmp);
|
|
current_input_set->set_row_mlpp_vector(j, row_tmp);
|
|
}
|
|
|
|
input_mini_batches.push_back(current_input_set);
|
|
}
|
|
|
|
/* Don't think this can ever happen, todo double check
|
|
if (real_t(n) / real_t(n_mini_batch) - int(n / n_mini_batch) != 0) {
|
|
for (int i = 0; i < n - n / n_mini_batch * n_mini_batch; i++) {
|
|
inputMiniBatches[n_mini_batch - 1].push_back(inputSet[n_mini_batch * n_mini_batch + i]);
|
|
}
|
|
}
|
|
*/
|
|
|
|
return input_mini_batches;
|
|
}
|
|
MLPPUtilities::CreateMiniBatchMVBatch MLPPUtilities::create_mini_batchesmv(const Ref<MLPPMatrix> &input_set, const Ref<MLPPVector> &output_set, int n_mini_batch) {
|
|
Size2i size = input_set->size();
|
|
|
|
int n = size.y;
|
|
int mini_batch_element_count = n / n_mini_batch;
|
|
|
|
Ref<MLPPVector> row_tmp;
|
|
row_tmp.instance();
|
|
row_tmp->resize(size.x);
|
|
|
|
CreateMiniBatchMVBatch ret;
|
|
|
|
for (int i = 0; i < n_mini_batch; i++) {
|
|
int mini_batch_start_offset = n_mini_batch * i;
|
|
Ref<MLPPMatrix> current_input_set;
|
|
current_input_set.instance();
|
|
current_input_set->resize(Size2i(size.x, mini_batch_element_count));
|
|
|
|
Ref<MLPPVector> current_output_set;
|
|
current_output_set.instance();
|
|
current_output_set->resize(mini_batch_element_count);
|
|
|
|
for (int j = 0; j < mini_batch_element_count; j++) {
|
|
int main_indx = mini_batch_start_offset + j;
|
|
|
|
input_set->get_row_into_mlpp_vector(main_indx, row_tmp);
|
|
current_input_set->set_row_mlpp_vector(j, row_tmp);
|
|
|
|
current_output_set->set_element(j, output_set->get_element(j));
|
|
}
|
|
|
|
ret.input_sets.push_back(current_input_set);
|
|
ret.output_sets.push_back(current_output_set);
|
|
}
|
|
|
|
/* Don't think this can ever happen, todo double check
|
|
if (real_t(n) / real_t(n_mini_batch) - int(n / n_mini_batch) != 0) {
|
|
for (int i = 0; i < n - n / n_mini_batch * n_mini_batch; i++) {
|
|
inputMiniBatches[n_mini_batch - 1].push_back(inputSet[n / n_mini_batch * n_mini_batch + i]);
|
|
outputMiniBatches[n_mini_batch - 1].push_back(outputSet[n / n_mini_batch * n_mini_batch + i]);
|
|
}
|
|
}
|
|
*/
|
|
|
|
return ret;
|
|
}
|
|
MLPPUtilities::CreateMiniBatchMMBatch MLPPUtilities::create_mini_batchesmm(const Ref<MLPPMatrix> &input_set, const Ref<MLPPMatrix> &output_set, int n_mini_batch) {
|
|
Size2i input_set_size = input_set->size();
|
|
Size2i output_set_size = output_set->size();
|
|
|
|
int n = input_set_size.y;
|
|
int mini_batch_element_count = n / n_mini_batch;
|
|
|
|
Ref<MLPPVector> input_row_tmp;
|
|
input_row_tmp.instance();
|
|
input_row_tmp->resize(input_set_size.x);
|
|
|
|
Ref<MLPPVector> output_row_tmp;
|
|
output_row_tmp.instance();
|
|
output_row_tmp->resize(output_set_size.x);
|
|
|
|
CreateMiniBatchMMBatch ret;
|
|
|
|
for (int i = 0; i < n_mini_batch; i++) {
|
|
int mini_batch_start_offset = n_mini_batch * i;
|
|
Ref<MLPPMatrix> current_input_set;
|
|
current_input_set.instance();
|
|
current_input_set->resize(Size2i(input_set_size.x, mini_batch_element_count));
|
|
|
|
Ref<MLPPMatrix> current_output_set;
|
|
current_output_set.instance();
|
|
current_output_set->resize(Size2i(output_set_size.x, mini_batch_element_count));
|
|
|
|
for (int j = 0; j < mini_batch_element_count; j++) {
|
|
int main_indx = mini_batch_start_offset + j;
|
|
|
|
input_set->get_row_into_mlpp_vector(main_indx, input_row_tmp);
|
|
current_input_set->set_row_mlpp_vector(j, input_row_tmp);
|
|
|
|
output_set->get_row_into_mlpp_vector(main_indx, output_row_tmp);
|
|
current_output_set->set_row_mlpp_vector(j, output_row_tmp);
|
|
}
|
|
|
|
ret.input_sets.push_back(current_input_set);
|
|
ret.output_sets.push_back(current_output_set);
|
|
}
|
|
|
|
/* Don't think this can ever happen, todo double check
|
|
if (real_t(n) / real_t(n_mini_batch) - int(n / n_mini_batch) != 0) {
|
|
for (int i = 0; i < n - n / n_mini_batch * n_mini_batch; i++) {
|
|
inputMiniBatches[n_mini_batch - 1].push_back(inputSet[n / n_mini_batch * n_mini_batch + i]);
|
|
}
|
|
}
|
|
*/
|
|
|
|
return ret;
|
|
}
|
|
|
|
Array MLPPUtilities::create_mini_batchesm_bind(const Ref<MLPPMatrix> &input_set, int n_mini_batch) {
|
|
Vector<Ref<MLPPMatrix>> batches = create_mini_batchesm(input_set, n_mini_batch);
|
|
|
|
Array ret;
|
|
|
|
for (int i = 0; i < batches.size(); ++i) {
|
|
ret.push_back(batches[i].get_ref_ptr());
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
Array MLPPUtilities::create_mini_batchesmv_bind(const Ref<MLPPMatrix> &input_set, const Ref<MLPPVector> &output_set, int n_mini_batch) {
|
|
CreateMiniBatchMVBatch batches = create_mini_batchesmv(input_set, output_set, n_mini_batch);
|
|
|
|
Array inputs;
|
|
Array outputs;
|
|
|
|
for (int i = 0; i < batches.input_sets.size(); ++i) {
|
|
inputs.push_back(batches.input_sets[i].get_ref_ptr());
|
|
outputs.push_back(batches.output_sets[i].get_ref_ptr());
|
|
}
|
|
|
|
Array ret;
|
|
|
|
ret.push_back(inputs);
|
|
ret.push_back(outputs);
|
|
|
|
return ret;
|
|
}
|
|
Array MLPPUtilities::create_mini_batchesmm_bind(const Ref<MLPPMatrix> &input_set, const Ref<MLPPMatrix> &output_set, int n_mini_batch) {
|
|
CreateMiniBatchMMBatch batches = create_mini_batchesmm(input_set, output_set, n_mini_batch);
|
|
|
|
Array inputs;
|
|
Array outputs;
|
|
|
|
for (int i = 0; i < batches.input_sets.size(); ++i) {
|
|
inputs.push_back(batches.input_sets[i].get_ref_ptr());
|
|
outputs.push_back(batches.output_sets[i].get_ref_ptr());
|
|
}
|
|
|
|
Array ret;
|
|
|
|
ret.push_back(inputs);
|
|
ret.push_back(outputs);
|
|
|
|
return ret;
|
|
}
|
|
|
|
std::tuple<real_t, real_t, real_t, real_t> MLPPUtilities::TF_PN(std::vector<real_t> y_hat, std::vector<real_t> y) {
|
|
real_t TP = 0;
|
|
real_t FP = 0;
|
|
real_t TN = 0;
|
|
real_t FN = 0;
|
|
for (uint32_t i = 0; i < y_hat.size(); i++) {
|
|
if (y_hat[i] == y[i]) {
|
|
if (y_hat[i] == 1) {
|
|
TP++;
|
|
} else {
|
|
TN++;
|
|
}
|
|
} else {
|
|
if (y_hat[i] == 1) {
|
|
FP++;
|
|
} else {
|
|
FN++;
|
|
}
|
|
}
|
|
}
|
|
return { TP, FP, TN, FN };
|
|
}
|
|
|
|
real_t MLPPUtilities::recall(std::vector<real_t> y_hat, std::vector<real_t> y) {
|
|
auto res = TF_PN(y_hat, y);
|
|
auto TP = std::get<0>(res);
|
|
//auto FP = std::get<1>(res);
|
|
//auto TN = std::get<2>(res);
|
|
auto FN = std::get<3>(res);
|
|
|
|
return TP / (TP + FN);
|
|
}
|
|
|
|
real_t MLPPUtilities::precision(std::vector<real_t> y_hat, std::vector<real_t> y) {
|
|
auto res = TF_PN(y_hat, y);
|
|
auto TP = std::get<0>(res);
|
|
auto FP = std::get<1>(res);
|
|
//auto TN = std::get<2>(res);
|
|
//auto FN = std::get<3>(res);
|
|
|
|
return TP / (TP + FP);
|
|
}
|
|
|
|
real_t MLPPUtilities::accuracy(std::vector<real_t> y_hat, std::vector<real_t> y) {
|
|
auto res = TF_PN(y_hat, y);
|
|
auto TP = std::get<0>(res);
|
|
auto FP = std::get<1>(res);
|
|
auto TN = std::get<2>(res);
|
|
auto FN = std::get<3>(res);
|
|
|
|
return (TP + TN) / (TP + FP + FN + TN);
|
|
}
|
|
real_t MLPPUtilities::f1_score(std::vector<real_t> y_hat, std::vector<real_t> y) {
|
|
return 2 * precision(y_hat, y) * recall(y_hat, y) / (precision(y_hat, y) + recall(y_hat, y));
|
|
}
|
|
|
|
void MLPPUtilities::_bind_methods() {
|
|
ClassDB::bind_method(D_METHOD("weight_initializationv", "weights", "type"), &MLPPUtilities::weight_initializationv, WEIGHT_DISTRIBUTION_TYPE_DEFAULT);
|
|
ClassDB::bind_method(D_METHOD("weight_initializationm", "weights", "type"), &MLPPUtilities::weight_initializationm, WEIGHT_DISTRIBUTION_TYPE_DEFAULT);
|
|
ClassDB::bind_method(D_METHOD("bias_initializationr"), &MLPPUtilities::bias_initializationr);
|
|
ClassDB::bind_method(D_METHOD("bias_initializationv", "z"), &MLPPUtilities::bias_initializationv);
|
|
|
|
ClassDB::bind_method(D_METHOD("performance_vec", "y_hat", "output_set"), &MLPPUtilities::performance_vec);
|
|
ClassDB::bind_method(D_METHOD("performance_mat", "y_hat", "y"), &MLPPUtilities::performance_mat);
|
|
ClassDB::bind_method(D_METHOD("performance_pool_int_array_vec", "y_hat", "output_set"), &MLPPUtilities::performance_pool_int_array_vec);
|
|
|
|
ClassDB::bind_method(D_METHOD("create_mini_batchesm", "input_set", "n_mini_batch"), &MLPPUtilities::create_mini_batchesm_bind);
|
|
ClassDB::bind_method(D_METHOD("create_mini_batchesmv", "input_set", "output_set", "n_mini_batch"), &MLPPUtilities::create_mini_batchesmv_bind);
|
|
ClassDB::bind_method(D_METHOD("create_mini_batchesmm", "input_set", "output_set", "n_mini_batch"), &MLPPUtilities::create_mini_batchesmm_bind);
|
|
|
|
BIND_ENUM_CONSTANT(WEIGHT_DISTRIBUTION_TYPE_DEFAULT);
|
|
BIND_ENUM_CONSTANT(WEIGHT_DISTRIBUTION_TYPE_XAVIER_NORMAL);
|
|
BIND_ENUM_CONSTANT(WEIGHT_DISTRIBUTION_TYPE_XAVIER_UNIFORM);
|
|
BIND_ENUM_CONSTANT(WEIGHT_DISTRIBUTION_TYPE_HE_NORMAL);
|
|
BIND_ENUM_CONSTANT(WEIGHT_DISTRIBUTION_TYPE_HE_UNIFORM);
|
|
BIND_ENUM_CONSTANT(WEIGHT_DISTRIBUTION_TYPE_LE_CUN_NORMAL);
|
|
BIND_ENUM_CONSTANT(WEIGHT_DISTRIBUTION_TYPE_LE_CUN_UNIFORM);
|
|
BIND_ENUM_CONSTANT(WEIGHT_DISTRIBUTION_TYPE_UNIFORM);
|
|
}
|