0

I am attempting to run the mnist example in matconvnet using parapooling. I have allocated 2 gpus, but it shows this error while trying to run cnn_mnist_experiments:

   Error using cnn_train>(spmd) (line 157)
Error detected on workers 1 2.

Error in cnn_train (line 157)
    spmd

Error in cnn_mnist (line 55)
[net, info] = trainfn(net, imdb, getBatch(opts), ...

Error in cnn_mnist_experiments (line 3)
[net_bn, info_bn] = cnn_mnist(...

Caused by:
    Error using ParameterServer/startWithMMap (line 170)
    An UndefinedFunction error was thrown on the workers for 'vl_cudatool'.  This may be because the
    file containing 'vl_cudatool' is not accessible on the workers.  Specify the required files for
    this parallel pool using the command: addAttachedFiles(pool, ...).  See the documentation for
    parpool for more details.
        Undefined function 'vl_cudatool' for input arguments of type 'single'.
    Error using ParameterServer/startWithMMap (line 170)
    An UndefinedFunction error was thrown on the workers for 'vl_cudatool'.  This may be because the
    file containing 'vl_cudatool' is not accessible on the workers.  Specify the required files for
    this parallel pool using the command: addAttachedFiles(pool, ...).  See the documentation for
    parpool for more details.
        Undefined function 'vl_cudatool' for input arguments of type 'single'.

Apparently, the workers cannot get access to "vl_cudatool" file. What could be the issue here?

talonmies
  • 70,661
  • 34
  • 192
  • 269
tdr
  • 11
  • 2

1 Answers1

0

Undefined function 'vl_cudatool' for input arguments of type 'single'.

It means that your vl_cudatool function is not in your path

try

which vl_cudatool

I don't know what vl_cudatool is. Maybe you can try some earlier version of MatConvNet?

DataHungry
  • 351
  • 2
  • 9
  • I tried that, but it showed 'not found'. I even looked at matconvnet's github repository for this function, but its not available as a .m file. – tdr Jun 22 '17 at 16:24
  • @tdr I don't know what vl_cudatool is. Maybe you can try some earlier version of MatConvNet? – DataHungry Jun 22 '17 at 17:45