adri-romsor / fitnets Goto Github PK

Hi, we try running your code. Everything runs fine but at line 111 in fitnets-training.py it gives this error
: 'VectorSpace' doesn't have an attribute 'shape'.

Any help regarding that would be appreciated.
Mayank

When running fitnets_training.py it show an error at line

from pylearn2 import train

"You are importing theano.sandbox.cuda. This is the old GPU back-end and "
unittest.case.SkipTest: You are importing theano.sandbox.cuda. This is the old GPU back-end and is removed from Theano. Use Theano 0.9 to use it. Even better, transition to the new GPU back-end! See https://github.com/Theano/Theano/wiki/Converting-to-the-new-gpu-back-end%28gpuarray%29

How can this be solved if I choose not to install Theano 0.9?

I just confused about how you update the lr during training. As mentioned in your paper, for cifar-10, the initial lr is set as 0.005. Does it mean that it will update during training or it is a fixed value?

In cifar10_fitnet19_all.yaml:

cost: !obj:costs.KDCost.KDCost {
teacher_path: "<path_teacher_pkl>",
temperature: 3,
lambda_teach: 4,
hints: [[10,1]],
},

what is the teacher_path supposed to be? When run, I get "No such file or directory: '<path_teacher_pkl>' ".

Also, What is the difference between cifar10_fitnet19_all.yaml and cifar10_fitnet19_valid.yam? What does the valid and all stand for?