improved performance

Author: dredstone1

include/network/INetwork.hpp

@@ -4,6 +4,7 @@
 #include "../src/model/optimizers.hpp"
 #include "IvisualNetwork.hpp"
 #include "tensor.hpp"
+#include <vector>
 
 namespace nn::model {
 class INetwork {
@@ -23,7 +24,7 @@ class INetwork {
 	virtual const global::Tensor &getInput() const = 0;
 	virtual std::shared_ptr<visualizer::IVisualNetwork> getVisual() = 0;
 
-	virtual global::Tensor getParams() const = 0;
+	virtual std::vector<global::ValueType> getParams() const = 0;
 	virtual void setParams(const global::Tensor params) = 0;
 
 	virtual void setTraining(const bool state) = 0;

src/model/model.cpp

@@ -1,10 +1,13 @@
 #include "../networks/cnn/CNNetwork.hpp"
 #include "../networks/fnn/FNNetwork.hpp"
 #include "dataBase.hpp"
+#include "tensor_gpu.hpp"
 #include <chrono>
 #include <fstream>
 #include <iostream>
 #include <model.hpp>
+#include <string>
+#include <vector>
 
 namespace nn::visualizer {
 void ProgressBar::printBar() {
@@ -104,7 +107,7 @@ void Model::initModel() {
 			addCNN(WIDTH, _config);
 		}
 
-		param_amount += network[i]->getParams().numElements();
+		// param_amount += network[i]->getParams().numElements();
 	}
 
 	std::cout << "initialize model - "
@@ -432,11 +435,11 @@ void Model::save(const std::string &file, bool print) {
 	}
 
 	for (size_t i = 0; i < network.size(); ++i) {
-		global::Tensor params = network[i]->getParams();
+        std::vector<global::ValueType> params = network[i]->getParams();
 
-		outFile << params.numElements() << " ";
-		for (size_t j = 0; j < params.numElements(); ++j) {
-			outFile << params.getValue({j}) << " ";
+		outFile << params.size() << " ";
+		for (size_t j = 0; j < params.size(); ++j) {
+			outFile << params[j] << " ";
 		}
 		outFile << std::endl;
 	}

src/networks/cnn/CNNetwork.cpp

@@ -1,6 +1,7 @@
 #include "CNNetwork.hpp"
 #include "tensor.hpp"
 #include <cstddef>
+#include <vector>
 
 namespace nn::model::cnn {
 CNNetwork::CNNetwork(
@@ -48,8 +49,8 @@ void CNNetwork::updateWeights(IOptimizer &) {
 void CNNetwork::calculateInputDelta(const global::Tensor &) {
 }
 
-global::Tensor CNNetwork::getParams() const {
-	return input;
+std::vector<global::ValueType> CNNetwork::getParams() const {
+	return std::vector<global::ValueType>();
 }
 
 void CNNetwork::setParams(const global::Tensor) {

src/networks/cnn/CNNetwork.hpp

@@ -5,6 +5,7 @@
 #include "tensor.hpp"
 #include <memory>
 #include <network/INetwork.hpp>
+#include <vector>
 
 namespace nn::model::cnn {
 class CNNetwork : public INetwork {
@@ -38,7 +39,7 @@ class CNNetwork : public INetwork {
 
 	std::shared_ptr<visualizer::IVisualNetwork> getVisual() override { return visual; }
 
-	global::Tensor getParams() const override;
+    std::vector<global::ValueType> getParams() const override;
 	void setParams(const global::Tensor params) override;
 
 	void setTraining(const bool state) override;

src/networks/fnn/DenseLayer.cpp

@@ -128,17 +128,28 @@ void DenseLayer::updateWeight(nn::model::IOptimizer &optimizer) {
 	optimizer.step(parameters.weights, gradients.weights);
 }
 
-const global::Tensor DenseLayer::getData() const {
+const std::vector<global::ValueType> DenseLayer::getData() const {
 	size_t weightsSize = parameters.weights.numElements();
 	size_t biasesSize = parameters.biases.numElements();
 
-	global::Tensor matrix({weightsSize + biasesSize});
+	std::vector<global::ValueType> matrix(weightsSize + biasesSize);
 
-	// Copy weights
-	matrix.insertRange(parameters.weights, 0, 0, weightsSize);
+	std::vector<global::ValueType> temp(parameters.weights.numElements());
+	parameters.weights.getData(temp);
+
+	size_t j = 0;
+	for (size_t i = 0; i < temp.size(); ++i) {
+		matrix[j] = temp[i];
+		++j;
+	}
 
-	// Copy biases
-	matrix.insertRange(parameters.biases, 0, weightsSize, biasesSize);
+	temp.resize(parameters.biases.numElements());
+	parameters.biases.getData(temp);
+
+	for (size_t i = 0; i < temp.size(); ++i) {
+		matrix[j] = temp[i];
+		++j;
+	}
 
 	return matrix;
 }

src/networks/fnn/DenseLayer.hpp

@@ -3,6 +3,7 @@
 
 #include "../src/model/optimizers.hpp"
 #include <Globals.hpp>
+#include <vector>
 
 namespace nn::model::fnn {
 constexpr global::ValueType MIN_LOSS_VALUE = 1e-10;
@@ -70,7 +71,7 @@ class DenseLayer {
 
 	size_t getParamCount() const;
 
-	const global::Tensor getData() const;
+	const std::vector<global::ValueType> getData() const;
 	void setData(const global::Tensor newParam, const size_t offset);
 
 	void setTraining(const bool state) { isTraining = state; }

src/networks/fnn/FNNetwork.cpp

@@ -1,5 +1,6 @@
 #include "FNNetwork.hpp"
 #include "tensor.hpp"
+#include <vector>
 
 namespace nn::model::fnn {
 FNNetwork::FNNetwork(
@@ -149,16 +150,18 @@ size_t FNNetwork::getParamCount() const {
 	return count;
 }
 
-global::Tensor FNNetwork::getParams() const {
-	global::Tensor matrix({getParamCount()});
+std::vector<global::ValueType> FNNetwork::getParams() const {
+	std::vector<global::ValueType> matrix(getParamCount());
 
 	size_t matrixI = 0;
 
 	for (size_t i = 0; i < layers.size(); ++i) {
-		global::Tensor params = layers[i]->getData();
+		std::vector<global::ValueType> params = layers[i]->getData();
 
-		matrix.insertRange(params, 0, matrixI, params.numElements());
-		matrixI += params.numElements();
+		for (size_t j = 0; j < params.size(); ++j) {
+			matrix[matrixI] = params[j];
+			++matrixI;
+		}
 	}
 
 	return matrix;
@@ -168,7 +171,7 @@ void FNNetwork::setParams(const global::Tensor params) {
 	size_t j = 0;
 	for (size_t i = 0; i < layers.size(); ++i) {
 		layers[i]->setData(params, j);
-        j += layers[i]->getParamCount();
+		j += layers[i]->getParamCount();
 
 		if (visual) {
 			visual->setParam(i, layers[i]->getParms());

src/networks/fnn/FNNetwork.hpp

@@ -3,6 +3,7 @@
 
 #include "FnnVisualizer.hpp"
 #include <network/INetwork.hpp>
+#include <vector>
 
 namespace nn::model::fnn {
 class FNNetwork : public INetwork {
@@ -44,7 +45,7 @@ class FNNetwork : public INetwork {
 		return visual;
 	}
 
-	global::Tensor getParams() const override;
+    std::vector<global::ValueType> getParams() const override;
 	void setParams(const global::Tensor params) override;
 
 	size_t getParamCount() const;