Questions and answers

You can submit your questions regarding Theano, Pylearn2, the class project or anything related as comments to this page.

20 thoughts on “Questions and answers”

  1. I have pylearn2 up and running on my laptop, but I get the following error when trying to train a model on my desktop:

    ImportError: Could not import pylearn2.models.mlp but could import pylearn2.models. Original exception: No module named dnn

    Original exception:
    ImportError: No module named dnn

    Has anyone else dealt with this? I’m on Windows 8.1 64-bit without an nVidia GPU

    Like

  2. This looks like a problem when pylearn2.models.mlp tries importing theano.sandbox.cuda.dnn. Can you pull the latest Theano changes and check again to see if that solves your issue?

    Like

  3. Hi Vincent,

    I have run several models with pylearn2 and saved the .pkl file each model

    Im trying to visualize the classification error etc… from that file but it do not seems to contain information on the training process except the model…
    like valid_y_misclass …
    Am I missing something? How to you extract that information to make plots? I saw there is a live monitor but I dont want to visualize the process during processing only make plots after.

    Thanks a lot

    Like

    1. To make plots you can use the pylearn2/pylearn2/scripts/plot_monitor.py script.
      You can run it in the bash and pass your .pkl file as parameter. You can also save your plot as a .png file (which is usefull when you’re in remote) by doing:
      plot_monitor.py –out=

      Liked by 1 person

        1. You can also look at the “monitor” attribute of the pickled model, which is the Monitor object containing all monitoring values computed during training.

          It has a “channels” attribute, which is a dict mapping channel names to MonitorChannel objects.

          A MonitorChannel object contains the monitoring values for a given monitoring channel. The values themselves are contained in the “val_record” attribute.

          Like

          1. I tried Safari, Firefox and Chrome, neither of them is feasible. Also, I can’t post comment on the new lecture page, even reply someone. I’ll sign up another account to try again.

            Like

          2. Vincent, I think I figured out the reason why I can’t post comments on page course. I added the link to a paper which may cause the issue, I changed the link now, and everything is okay! Anyway, thanks for your help!

            Like

  4. I’m getting an error when I try to set the scaled_size and crop_size of the transformer to anything other than 256 and 221

    ERROR (theano.gof.opt): Optimization failure due to: local_gpu_downsample_factor_max
    ERROR:theano.gof.opt:Optimization failure due to: local_gpu_downsample_factor_max

    I tried adding some flags to the blas library like blow as suggested by a stackoverflow post but it didn’t help 😦 Does anyone have any other ideas?

    [blas]
    ldflags=-lblas -lgfortran

    Like

  5. Hi, I have a question about the monitoring channels in Pylearn2. For a layer ‘h1’, I have a lot of channels ‘train_h1_max_x_mean_u’, ‘train_h1_max_x_min_u’… I’m confused as to what they mean. For ‘train_h1_max_x_mean_u’… How is max_x determined? The maximum pre-activation seen from all training examples? Is it the maximum of the mean of all the neurons?

    Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s