Tianxu Jia
Dear All
Now I am using C++ for training Lenet5 save it and load it again and testing it. I don’t know whether I am doing it right.
My code for training and saving it
#include "stdafx.h"
#include "libeblearn.h"
#include "libeblearngui.h"
#include "idxIO.h"
#include "libidxgui.h"
#ifdef _DEBUG
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\debug\\eblearntools.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\debug\\eblearn.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\debug\\idx.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\debug\\idxgui.lib");
#pragma comment(lib,"C:\\QtSDK\\Desktop\\Qt\\4.8.1\\msvc2010\\lib\\QtGuid4.lib");
#pragma comment(lib,"C:\\QtSDK\\Desktop\\Qt\\4.8.1\\msvc2010\\lib\\QtCored4.lib");
#else
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\Release\\eblearntools.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\Release\\eblearn.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\Release\\idx.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\Release\\idxgui.lib");
#endif
MAIN_QTHREAD(int, argc, char**, argv)
{
typedef double Tnet;
typedef ebl::ubyte Tdata;
typedef ebl::ubyte Tlab;
bool display = true;
uint ninternals = 1;
ebl::init_drand(time(NULL)); // initialize random seed
std::string mnist_dir = "D:\\work\\EBLearn\\eblearn-1.0\\data";
//! load MNIST datasets
ebl::mnist_datasource<Tnet, ebl::ubyte,ebl::ubyte>
train_ds(mnist_dir.c_str(), true, 2000);
train_ds.set_balanced(true);
train_ds.set_shuffle_passes(true);
train_ds.set_weigh_samples(false, false, false, 0.01);
train_ds.set_epoch_show(100); // show progress every 100 samples
train_ds.ignore_correct(true);
// create 1-of-n targets with target 1.0 for shown class, -1.0 for the rest
ebl::intg temp = 1+idx_max(train_ds.labels);
ebl::idx<Tnet> targets =
ebl::create_target_matrix<Tnet>(1+idx_max(train_ds.labels), 1.0);
uint nclasses = targets.dim(0);
// create the network weights, network and trainer
ebl::parameter<Tnet> theparam(60000); // create trainable parameter
ebl::lenet5<Tnet> net(theparam, 32, 32, 5, 5, 2, 2, 5, 5, 2, 2, 120, nclasses, true, false, true, false);
std::cout << net.describe() << std::endl;
ebl::l2_energy<Tnet> energy;
ebl::class_answer<Tnet,Tdata,Tlab> answer(nclasses);
ebl::trainable_module<Tnet,Tdata,Tlab> trainable(energy, net, NULL, NULL, &answer);
ebl::supervised_trainer<Tnet, ebl::ubyte, ebl::ubyte> thetrainer(trainable, theparam);
ebl::supervised_trainer_gui<Tnet,Tdata,Tlab> stgui;
// a classifier-meter measures classification errors
ebl::classifier_meter trainmeter;
// initialize the network weights
ebl::forget_param_linear fgp(1, 0.5);
trainable.forget(fgp);
// gradient parameters
ebl::gd_param gdp(/* double eta*/ 0.0001,
/* double ln */ 0.0,
/* double l1 */ 0.0,
/* double l2 */ 0.0,
/* intg dtime */ 0,
/* double iner */0.0,
/* double anneal_value */ 0.001,
/* intg anneal_period */ 2000,
/* double g_t*/ 0.0);
std::cout << "gdp = "<<gdp << std::endl;
ebl::infer_param infp;
ebl::fixed_init_drand(); // fixed randomization
thetrainer.compute_diaghessian(train_ds, 100, 0.02);
for (int i = 0; i < 5; ++i) {
thetrainer.train(train_ds, trainmeter, gdp, 1, infp);
thetrainer.compute_diaghessian(train_ds, 100, 0.02);
}
const std::string filename = "D:\\work\\EBLearn\\eblearn-1.0\\lenet5.mat";
theparam.save_x(filename.c_str());
std::cout<<"OK";
return 0;
}
Here is my code of loading the lenet5 machine
#include "stdafx.h"
#include "libeblearn.h"
#include "libeblearngui.h"
#include "libidxgui.h"
#ifdef _DEBUG
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\debug\\eblearntools.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\debug\\eblearn.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\debug\\idx.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\debug\\idxgui.lib");
#pragma comment(lib,"C:\\QtSDK\\Desktop\\Qt\\4.8.1\\msvc2010\\lib\\QtGuid4.lib");
#pragma comment(lib,"C:\\QtSDK\\Desktop\\Qt\\4.8.1\\msvc2010\\lib\\QtCored4.lib");
#else
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\Release\\eblearntools.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\Release\\eblearn.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\Release\\idx.lib");
#pragma comment(lib,"D:\\work\\EBLearn\\eblearn-1.0\\bin\\Release\\idxgui.lib");
#endif
MAIN_QTHREAD(int, argc, char**, argv)
{
typedef double Tnet;
typedef ebl::ubyte Tdata;
typedef ebl::ubyte Tlab;
bool display = true;
uint ninternals = 1;
ebl::init_drand(time(NULL)); // initialize random seed
std::string mnist_dir = "D:\\work\\EBLearn\\eblearn-1.0\\data";
//! load MNIST datasets: trsize for training set and tesize for testing set
ebl::mnist_datasource<Tnet, ebl::ubyte,ebl::ubyte>
test_ds(mnist_dir.c_str(), false, 1000);
// create 1-of-n targets with target 1.0 for shown class, -1.0 for the rest
ebl::idx<Tnet> targets =
ebl::create_target_matrix<Tnet>(1+idx_max(test_ds.labels), 1.0);
uint nclasses = targets.dim(0);
// create the network weights, network and trainer
ebl::idxdim dims(test_ds.sample_dims()); // get order and dimensions of sample
ebl::parameter<Tnet> theparam;// create trainable parameter
std::string file = "D:\\work\\EBLearn\\eblearn-1.0\\lenet5.mat";
theparam.load_x(file.c_str());
ebl::lenet5<Tnet> net(theparam, 32, 32, 5, 5, 2, 2, 5, 5, 2, 2, 120, nclasses, true, false, true, false);
return 0;
}
Here are my problems:
1. Is my way of saving the trained machine(lenet5) correct? What is the right way of doing it if it is wrong?
2. Is my way of using the trained machine saved in file correct? Otherwise, how to do it?
3. I don’t know the API of running(testing) the machine(lenet5) with input a frame of image. Can you please tell me how to do?
4. My code always has run time exception at the line
theparam.load_x(file.c_str());
I don’t know why
I am struggling with these problems. Can you so kindly give me a help? It will be great help if you can show me with several pieces of codes
You any suggestion will be greatly appreciated
Best regards
Sunny