Ejemplo n.º 1
0
        static void Main(string[] args)
        {
            string filename = "DBN-";

            filename += DateTime.Now.Ticks;
            filename += ".txt";
            StreamWriter sw = new StreamWriter(filename);

            Console.SetOut(sw);
            sw.AutoFlush = true;

            MnistDataMgr trainDataMgr = new MnistDataMgr("train.csv");

            trainDataMgr.Load();

            Console.WriteLine("Train data loaded");

            MnistDataMgr testDataMgr = new MnistDataMgr("test.csv");

            testDataMgr.Load();

            Console.WriteLine("Test data loaded");

            //int [] netStruct = {trainDataMgr.inputNum, 100, 50, 20};
            //int[] netStruct = { trainDataMgr.inputNum, 200 };
            int[] netStruct = { trainDataMgr.inputNum, 200 };


            LRBM lrbm = new LRBM(1, netStruct);

            LRBMTrainer lrbmTrainer = new LRBMTrainer(trainDataMgr, lrbm);

            lrbmTrainer.BatchTrain();
            //lrbmTrainer.OnlineTrain();

            lrbm.DumpToFile();

            //DeepBeliefNet dbn = new DeepBeliefNet(0.8, lrbm, 10);
            DeepBeliefNet dbn = new DeepBeliefNet(0.8, lrbm, new int[] { 200, 10 });

            dbn.DumpToFile();
            dbn.GetMLP().DumpToFile();
            //dbn.GetMLP().ClearBias();

            //dbn.GetMLP()._trainType = MLP.TrainType.TRAIN_ONLY_OUTPUT_LAYER;
            dbn.GetMLP()._trainType    = MLP.TrainType.NORMAL_BP;
            dbn.GetMLP().learningDepth = 2;

            MnistMLPTrainer trainer = new MnistMLPTrainer(trainDataMgr, dbn.GetMLP());

            trainer.OnlineTrain();

            MnistMLPTester tester = new MnistMLPTester(dbn.GetMLP(), testDataMgr);

            tester.Test();

            sw.Flush();
            sw.Close();
        }
Ejemplo n.º 2
0
        static void Main(string[] args)
        {
            string filename = "DBN-";

            filename += DateTime.Now.Ticks;
            filename += ".txt";
            StreamWriter sw = new StreamWriter(filename);

            Console.SetOut(sw);
            sw.AutoFlush = true;

            MnistDataMgr trainDataMgr = new MnistDataMgr("train.csv");

            trainDataMgr.Load();

            Console.WriteLine("Train data loaded");

            MnistDataMgr testDataMgr = new MnistDataMgr("test.csv");

            testDataMgr.Load();

            Console.WriteLine("Test data loaded");


            /*
             * Stream stream2 = File.Open("lrbm-param.osl", FileMode.Open);
             * BinaryFormatter bformatter2 = new BinaryFormatter();
             *
             * Console.WriteLine("Reading Employee Information");
             * LRBM lrbm2 = (LRBM)bformatter2.Deserialize(stream2);
             * stream2.Close();
             *
             * //lrbm2.DumpToFile();
             *
             *
             *
             * //dbn2.DumpToFile();
             * //dbn2.GetMLP().DumpToFile();
             *
             * //dbn2.GetMLP()._trainType = MLP.TrainType.NORMAL_BP;
             *
             * //MnistMLPTrainer trainer = new MnistMLPTrainer(trainDataMgr, dbn2.GetMLP());
             * //trainer.OnlineTrain();
             *
             * //MnistMLPTester tester = new MnistMLPTester(dbn2.GetMLP(), testDataMgr);
             * //tester.Test();
             *
             * return;
             */



            //int [] netStruct = {trainDataMgr.inputNum, 100, 50, 20};
            int[] netStruct = { trainDataMgr.inputNum, 200 };
            //int[] netStruct = { trainDataMgr.inputNum, 100 };

            /*
             * RBM rbm = new RBM(0.2, trainDataMgr.inputNum, 100);
             *
             * RBMTrainer rbmTrainer = new RBMTrainer(trainDataMgr, rbm);
             * //rbmTrainer.OnlineTrain();
             * rbmTrainer.BatchTrain();
             *
             * rbm.DumpToFile();
             *
             * rbm.DumpParamsToFile("rbmParams.osl");
             *
             * int testSize = 40;
             *
             * for (int t = 0; t < testSize; t++)
             * {
             *   Random rnd1 = new Random();
             *   int fileIndex = rnd1.Next(trainDataMgr.count-1);
             *   double[] sample = trainDataMgr.GetNormalizedInputData(fileIndex);
             *
             *   string wrFilename = "sample-input-" + fileIndex + ".csv";
             *
             *   StreamWriter sampleWr = new StreamWriter(wrFilename);
             *   for (int i = 0; i < sample.Length; i++)
             *   {
             *       sampleWr.Write(sample[i] + ", ");
             *   }
             *
             *   sampleWr.Flush();
             *   sampleWr.Close();
             *
             *   rbm.Calc(sample);
             *   double[] sampleOut = rbm.GetVisible();
             *   double[] sampleOutProb = rbm.GetVisibleProb();
             *
             *   wrFilename = "sample-prob-output-" + fileIndex + ".csv";
             *
             *   StreamWriter outProbWr = new StreamWriter(wrFilename);
             *   for (int i = 0; i < sampleOutProb.Length; i++)
             *   {
             *       outProbWr.Write(sampleOutProb[i] + ", ");
             *   }
             *
             *   outProbWr.Flush();
             *   outProbWr.Close();
             *
             *   wrFilename = "sample-output-" + fileIndex + ".csv";
             *   StreamWriter outWr = new StreamWriter(wrFilename);
             *   for (int i = 0; i < sampleOut.Length; i++)
             *   {
             *       outWr.Write(sampleOut[i] + ", ");
             *   }
             *
             *   outWr.Flush();
             *   outWr.Close();
             * }
             *
             * return;
             *
             */


            LRBM lrbm = new LRBM(1, netStruct);

            LRBMTrainer lrbmTrainer = new LRBMTrainer(trainDataMgr, lrbm);

            lrbmTrainer.BatchTrain();
            //lrbmTrainer.OnlineTrain();

            lrbm.DumpToFile();
            lrbm.DumpParamsToFile("lrbm-param.osl");

            return;

            DeepBeliefNet dbn = new DeepBeliefNet(0.8, lrbm, 10);

            //DeepBeliefNet dbn = new DeepBeliefNet(0.8, lrbm, new int[]{50,10});
            dbn.DumpToFile();
            dbn.GetMLP().DumpToFile();

            dbn.DumpParamsToFile("dbn-ser.osl");

            return;



            //dbn.GetMLP().DumpToFile();

            dbn.GetMLP()._trainType = MLP.TrainType.TRAIN_ONLY_OUTPUT_LAYER;
            //dbn.GetMLP()._trainType = MLP.TrainType.NORMAL_BP;

            //MnistMLPTrainer trainer = new MnistMLPTrainer(trainDataMgr, dbn.GetMLP());
            //trainer.OnlineTrain();

            //MnistMLPTester tester = new MnistMLPTester(dbn.GetMLP(), testDataMgr);
            //tester.Test();

            //dbn.GetMLP().DumpToFile();

            sw.Flush();
            sw.Close();
        }