Errors compiling the mnist cpp sample code.

[ZachTM]

Hi again.  I am trying to figure out how to get eblearn to work in C++, to classify digits.  I have taken this code found in the documentation for a sample C++ usage:

#include "libeblearn.h"

#include <iostream>

#ifdef __GUI__
#include "libeblearngui.h"
#endif

using namespace std;
using namespace ebl; // all eblearn objects are under the ebl namespace

// argv[1] is expected to contain the directory of the mnist dataset
#ifdef __GUI__
MAIN_QTHREAD() { // this is the macro replacing main to enable multithreaded gui
#else
int main(int argc, char **argv) { // regular main without gui
#endif
  cout << "* MNIST demo: learning handwritten digits using the eblearn";
  cout << " C++ library *" << endl;
  if (argc != 2) {
    cout << "Usage: ./mnist <my mnist directory>" << endl;
    eblerror("MNIST path not specified");
  }
  init_drand(time(NULL)); // initialize random seed

  intg trsize = 60000; // maximum training set size: 60000
  intg tesize = 10000; // maximum testing set size:  10000

  //! load MNIST datasets: trize for training set and tesize for testing set
  mnist_datasource<ubyte,ubyte> train_ds, test_ds;
  load_mnist_dataset(argv[1], train_ds, test_ds, trsize, tesize);

  //! create 1-of-n targets with target 1.0 for shown class, -1.0 for the rest
  idx<double> targets = create_target_matrix(1+idx_max(train_ds.labels), 1.0);

  //! create the network weights, network and trainer
  idxdim dims(train_ds.sample_dims()); // get order and dimensions of sample
  parameter theparam(60000); // create trainable parameter
  lenet5 l5(theparam, 32, 32, 5, 5, 2, 2, 5, 5, 2, 2, 120, targets.dim(0));
  supervised_euclidean_machine thenet(l5, targets, dims);
  supervised_trainer<ubyte,ubyte> thetrainer(thenet, theparam);
  supervised_trainer_gui stgui; // the gui to display supervised_trainer

  //! a classifier-meter measures classification errors
  classifier_meter trainmeter, testmeter;

  //! initialize the network weights
  forget_param_linear fgp(1, 0.5);
  thenet.forget(fgp);

  // learning parameters
  gd_param gdp(/* double leta*/ 0.0001,
           /* double ln */     0.0,
           /* double l1 */     0.0,
           /* double l2 */     0.0,
           /* int dtime */     0,
            /* double iner */0.0,
           /* double a_v */ 0.0,
            /* double a_t */ 0.0,
           /* double g_t*/     0.0);
 infer_param infp;

  // estimate second derivative on 100 iterations, using mu=0.02
   cout << "Computing second derivatives on MNIST dataset: ";
   thetrainer.compute_diaghessian(train_ds, 100, 0.02);

  // first show classification results without training
   thetrainer.test(train_ds, trainmeter, infp);
   thetrainer.test(test_ds, testmeter, infp);
   stgui.display_datasource(thetrainer, test_ds, infp, 10, 10);
   stgui.display_internals(thetrainer, test_ds, infp, 2);

   // now do training iterations
   cout << "Training network on MNIST with " << train_ds.size();
   cout << " training samples and " << test_ds.size() << " test samples:" << endl;
   for (int i = 0; i < 100; ++i) {
     thetrainer.train(train_ds, trainmeter, gdp, 1);             // train
     thetrainer.test(train_ds, trainmeter, infp);             // test
     thetrainer.test(test_ds, testmeter, infp);                     // test
    stgui.display_datasource(thetrainer, test_ds, infp, 10, 10); // display
     stgui.display_internals(thetrainer, test_ds, infp, 2);       // display
     thetrainer.compute_diaghessian(train_ds, 100, 0.02); // recompute 2nd der
   }
   return 0;
}

I have added /usr/include/eblearn to my include paths, and  /usr/lib to my library search path with idx, idxgui, eblearngui, and eblearn all added to the list of libraries in my eclipse CDT project settings.  All the includes, and the namespace usage of ebl work fine with no errors. 

But then i start getting errors starting with :

   mnist_datasource<ubyte,ubyte> train_ds, test_ds;

saying: " invalid type in declaration before ‘,’ token"

and many other errors that all have something to do with undeclared functions or types and things like that.  Does anyone know what else I need to include to get this sample code running properly?  I would really appreciate any suggestions!  Thanks!

[Pierre Sermanet]

Hey,

This code is probably outdated, your best bet is to start from the train.cpp code.

Pierre

--
 
 

[ZachTM]

Ok thank you!