[Rbm]

- added JSON
- refactored


git-svn-id: http://moon:8086/svn/software/trunk/projects/Rbm@574 b431acfa-c32f-4a4a-93f1-934dc6c82436

Makefile

@@ -1,5 +1,5 @@
 CONFIG ?= release
-SRCS := source/main.cpp source/Rbm.cpp source/noise.c
+SRCS := source/main.cpp source/Rbm.cpp source/Layer.cpp source/noise.c
 	
 	
 LIBS := -larmadillo -ljsoncpp

source/Rbm.cpp

@@ -14,16 +14,13 @@
 #include "Rbm.hpp"
 #include "noise.h"
 
-Rbm::Rbm(const Params& params, arma::mat &w, arma::mat &bv, arma::mat &bh)
+Rbm::Rbm(const Params& params, size_t numVisible, size_t numHidden)
 : m_params(params)
-, m_w(w)
+, m_w(numVisible, numHidden)
-, m_bv(bv)
+, m_bv(1, numVisible)
-, m_bh(bh)
+, m_bh(1, numHidden)
 {
-	Noise_Init(&m_noise, 0x32727155);
+	weightsInit(0.0, m_params.weightInit);
-	uniform(m_w, 0.0, m_params.weightInit);
-	uniform(m_bh, 0.0, m_params.weightInit);
-	uniform(m_bv, 0.0, m_params.weightInit);
 }
 
 Rbm::Rbm(const Rbm& orig)
@@ -38,6 +35,28 @@ Rbm::~Rbm()
 {
 }
 
+void Rbm::weightsInit(double mu, double stddev)
+{
+	uniform(m_w, mu, stddev);
+	uniform(m_bh, mu, stddev);
+	uniform(m_bv, mu, stddev);
+
+}
+
+void Rbm::fromJson(Json::Value params)
+{
+}
+
+Json::Value Rbm::toJson() const
+{
+	Json::Value rbm;
+	rbm["numVisible"] = m_bv.n_elem;
+	rbm["numHidden"] = m_bh.n_elem;
+	rbm["params"] = m_params.toJson();
+	return rbm;
+}
+
+
 void Rbm::train(const arma::mat& batch, size_t miniBatchSize, size_t numEpochs, IListener* pListener)
 {
 	size_t epoch;
@@ -188,43 +207,56 @@ arma::mat Rbm::probsLogistic(const arma::mat &src)
 arma::mat Rbm::sample(const arma::mat &src)
 {
 	arma::mat dst = src;
+	uniform(dst);
+	
 	for (size_t i=0; i < src.n_rows; i++)
 	{
 		for (size_t j=0; j < src.n_cols; j++)
 		{
-			dst(i, j) = src(i, j) >= Noise_Uniform(&m_noise);
+			dst(i, j) = src(i, j) >= dst(i, j);
 		}
 	}
 	return dst;
 }
 
-arma::mat Rbm::toHiddenState(const arma::mat &visible)
+arma::mat Rbm::toHiddenState(const arma::mat &visible) const
 {
 	return visible * m_w  + arma::repmat(m_bh, visible.n_rows, 1);
 }
 
-arma::mat Rbm::toVisibleState(const arma::mat &hidden)
+arma::mat Rbm::toVisibleState(const arma::mat &hidden) const
 {
 	return hidden * m_w.t() + arma::repmat(m_bv, hidden.n_rows, 1);
 }
 
-arma::mat Rbm::toHiddenProbs(const arma::mat &visible)
+arma::mat Rbm::toHiddenProbs(const arma::mat &visible) const
 {
 	return probsLogistic(toHiddenState(visible));
 }
 
-arma::mat Rbm::toVisibleProbs(const arma::mat &hidden)
+arma::mat Rbm::toVisibleProbs(const arma::mat &hidden) const
 {
 	return probsLogistic(toVisibleState(hidden));
 }
 
 void Rbm::uniform(arma::mat& srcDst, double mu, double stdDev)
 {
-	for (size_t i=0; i < srcDst.n_rows; i++)
+	srcDst = stdDev*arma::randu(srcDst.n_rows, srcDst.n_cols) + mu;
-	{
-		for (size_t j=0; j < srcDst.n_cols; j++)
-		{
-			srcDst(i, j) = stdDev*Noise_Uniform(&m_noise) + mu;
-		}
-	}
 }
+
+const arma::mat& Rbm::w() const
+{
+	return m_w;
+}
+
+const arma::mat& Rbm::bv() const
+{
+	return m_bv;
+}
+
+const arma::mat& Rbm::bh() const
+{
+	return m_bh;
+}
+
+

source/Rbm.hpp

@@ -15,7 +15,7 @@
 #define RBM_HPP
 
 #include <armadillo>
-#include "noise.h"
+#include <jsoncpp/json/json.h>
 
 class Rbm
 {
@@ -25,7 +25,7 @@ public:
 	{
 		Params()
 		: weightInit(0.01)
-		, weightDecay(0.001)
+		, weightDecay(0.0)
 		, learningRate(0.1)
 		, momentum(0.5)
 		, doRaoBlackwell(true)
@@ -36,6 +36,35 @@ public:
 		{
 		}
 
+		Json::Value toJson() const
+		{
+			Json::Value params;
+			params["weightInit"] = weightInit;
+			params["weightDecay"] = weightDecay;
+			params["learningRate"] = learningRate;
+			params["momentum"] = momentum;
+			params["doRaoBlackwell"] = (int)doRaoBlackwell;
+			params["gibbsDoSampleVisible"] = (int)gibbsDoSampleVisible;
+			params["gibbsDoSampleHidden"] = (int)gibbsDoSampleHidden;
+			params["doSampleBatch"] = (int)doSampleBatch;
+			params["numGibbs"] = (int)numGibbs;
+
+			return params;
+		}
+
+		void fromJson(Json::Value params)
+		{
+			weightInit = params["weightInit"].asDouble();
+			weightDecay = params["weightDecay"].asDouble();
+			learningRate = params["learningRate"].asDouble();
+			momentum = params["momentum"].asDouble();
+			doRaoBlackwell = params["doRaoBlackwell"] == 1;
+			gibbsDoSampleVisible = params["gibbsDoSampleVisible"] == 1;
+			gibbsDoSampleHidden = params["gibbsDoSampleHidden"] == 1;
+			doSampleBatch = params["doSampleBatch"] == 1;
+			numGibbs = params["numGibbs"].asUInt();
+		}
+		
 		double weightInit;
 		double weightDecay;
 		double learningRate;
@@ -70,26 +99,29 @@ public:
 		}
 	};
 
-	Rbm(const Params& params, arma::mat &w, arma::mat &bv, arma::mat &bh);
+	Rbm(const Params& params, size_t numVisible, size_t numHidden);
 	Rbm(const Rbm& orig);
 	virtual ~Rbm();
 
+	void weightsInit(double mu, double stddev);
 	void train(arma::mat const &batch, size_t miniBatchSize, size_t numEpochs, IListener *pListener);
 	
-	arma::mat toHiddenState(const arma::mat &visible);
+	arma::mat toHiddenState(const arma::mat &visible) const;
-	arma::mat toVisibleState(const arma::mat &hidden);
+	arma::mat toVisibleState(const arma::mat &hidden) const;
-	arma::mat toHiddenProbs(const arma::mat &visible);
+	arma::mat toHiddenProbs(const arma::mat &visible) const;
-	arma::mat toVisibleProbs(const arma::mat &hidden);
+	arma::mat toVisibleProbs(const arma::mat &hidden) const;
-	arma::mat& weights();
+	const arma::mat& w() const;
-	arma::mat& bias_visible();
+	const arma::mat& bv() const;
-	arma::mat& bias_hidden();
+	const arma::mat& bh() const;
+	
+	Json::Value toJson() const;
+	void fromJson(Json::Value params);
 	
 private:
-	noise_gen_t m_noise;
 	const Params &m_params;
-	arma::mat &m_w;
+	arma::mat m_w;
-	arma::mat &m_bh;
+	arma::mat m_bh;
-	arma::mat &m_bv;
+	arma::mat m_bv;
 	arma::mat sample(arma::mat const &src);
 	static arma::mat probsLogistic(arma::mat const &src);
 	void uniform(arma::mat &srcDst, double mu=0.0, double stdDev=1.0);

source/main.cpp

@@ -7,6 +7,7 @@
 #include <armadillo>
 #include <jsoncpp/json/json.h>
 #include "Rbm.hpp"
+#include "Layer.hpp"
 
 using namespace std;
 class RbmListener : public Rbm::IListener
@@ -111,51 +112,22 @@ void saveWeight(const string &filename, size_t numVisibleX, size_t numVisibleY,
 	fclose(pFile);
 }
 
-void saveProject(const string &prjname)
+void saveProject(const string &prjname, const Layer &layer, size_t numTraining)
 {
 	ofstream ofs(prjname + string(".prj"));
 	
 	Json::StyledWriter writer;
 	Json::Value project;
 	project["project"]["name"] = prjname;
+	project["project"]["num_training"] = (int)numTraining;
+	project["project"]["training_file"] = prjname + string(".training.dat");
 	
-	Json::Value layer1;
+	Json::Value jsonLayer = layer.toJson();
-	layer1["name"] = "1";
+
-	layer1["num_hidden"] = 64;
+	Json::Value jsonLayers(Json::arrayValue);
-	layer1["num_visible"] = 28*28;
+	jsonLayers.append(jsonLayer);
-	Json::Value params1;
-	params1["weightInit"] = 0.01;
-	params1["weightDecay"] = 0.001;
-	params1["learningRate"] = 0.1;
-	params1["momentum"] = 0.5;
-	params1["doRaoBlackwell"] = 1;
-	params1["gibbsDoSampleVisible"] = 0;
-	params1["gibbsDoSampleHidden"] = 1;
-	params1["doSampleBatch"] = 0;
-	params1["numGibbs"] = 1;
-	layer1["params"] = params1;
 	
-	Json::Value layer2;
+	project["project"]["layers"] = jsonLayers;
-	layer2["name"] = "2";
-	layer2["num_hidden"] = 16;
-	layer2["num_visible"] = 64;
-	Json::Value params2;
-	params2["weightInit"] = 0.01;
-	params2["weightDecay"] = 0.001;
-	params2["learningRate"] = 0.1;
-	params2["momentum"] = 0.5;
-	params2["doRaoBlackwell"] = 1;
-	params2["gibbsDoSampleVisible"] = 0;
-	params2["gibbsDoSampleHidden"] = 1;
-	params2["doSampleBatch"] = 0;
-	params2["numGibbs"] = 1;
-	layer2["params"] = params2;
-	
-	Json::Value layers(Json::arrayValue);
-	layers.append(layer1);
-	layers.append(layer2);
-	
-	project["project"]["layers"] = layers;
 	
 	ofs << writer.write(project);
 }
@@ -164,25 +136,29 @@ int main()
 {
 	printf("Hallo, Welt!\n");
 
-	const string project("mnist_2");
+	const string project("mnist");
-	saveProject(project);
 	
 	RbmListener statusDisplay;
-	Rbm::Params params;
+	Rbm::Params rbmParams;
+
 	arma::mat batch = loadTraining(project + string(".training.dat"));
 
 	size_t numTraining = batch.n_rows;
 	size_t numVisible = batch.n_cols;
 	size_t numHidden = 64;
 
+	Layer layer(project, 0, numVisible, numHidden);
+	
+	saveProject(project, layer, numTraining);
+
 	printf("Loaded %d training samples\n", (int)numTraining);
 
 	arma::mat w = arma::zeros(numVisible, numHidden);
 	arma::mat bv = arma::zeros(1, numVisible);
 	arma::mat bh = arma::zeros(1, numHidden);
-	Rbm rbm(params, w, bv, bh);
+	Rbm rbm(rbmParams, numVisible, numHidden);
 	rbm.train(batch, 1000, 100, &statusDisplay);
-	saveWeight(project + string(".weights.dat"), 28, 28, w, bv, bh);
+	saveWeight(project + string(".weights.dat"), 28, 28, rbm.w(), rbm.bv(), rbm.bh());
 	arma::mat v = arma::randu(numTraining, numVisible);
 	arma::mat h = rbm.toHiddenProbs(v);
 	arma::mat r = rbm.toVisibleProbs(h);