Exception in mnist_cnn.cpp

[AdmiralPellaeon]

I build mlpack with VS2019 x64 for Windows. I also created the needed CSV file with the data with the help of the python script from the tools directory.

I tried to run the "mnist_cnn.cpp" from this repo. Compiling works fine, but during the training an exception is thrown: "error: Mat::operator(): index out of bounds".

This is the call stack until the exception occurs:

Best regards

[rcurtin]

Hey @AdmiralPellaeon, thanks for the report. What version of mlpack are you compiling against?

[AdmiralPellaeon]

It's mlpack-3.4.2.

[rcurtin]

Ah, ok, thanks for the information. I think that is the issue. mlpack-3.4.2 is pretty old now but we haven't released mlpack 4 yet because we are in the midst of a big refactoring. If you can download the latest git master branch from https://github.com/mlpack/mlpack and use that and compile the MNIST CNN example against that version, it should work. If you can't, then I think this should fix it, if you change these lines:

  // Create labels for training and validatiion datasets.
  const mat trainY = train.row(0);
  const mat validY = valid.row(0);

to this:

  // Create labels for training and validatiion datasets.
  const mat trainY = train.row(0) + 1;
  const mat validY = valid.row(0) + 1;

Sorry for the inconvenience! I hope we will be able to release and package a new version of mlpack soon (there is still lots of refactoring to do though).

[AdmiralPellaeon]

Thanks for the answer. I will test it!

[dfy888]

 train.row(0)

Ah, ok, thanks for the information. I think that is the issue. mlpack-3.4.2 is pretty old now but we haven't released mlpack 4 yet because we are in the midst of a big refactoring. If you can download the latest git master branch from https://github.com/mlpack/mlpack and use that and compile the MNIST CNN example against that version, it should work. If you can't, then I think this should fix it, if you change these lines:

  // Create labels for training and validatiion datasets.
  const mat trainY = train.row(0);
  const mat validY = valid.row(0);

to this:

  // Create labels for training and validatiion datasets.
  const mat trainY = train.row(0) + 1;
  const mat validY = valid.row(0) + 1;

Sorry for the inconvenience! I hope we will be able to release and package a new version of mlpack soon (there is still lots of refactoring to do though).

yes,it is work!