Make all tutorials into unit tests

[lamblin]

All tutorials in pylearn2/scripts/tutorials should not get broken, and be updated when the interface changes or things get deprecated.

Here is the status of current tutorials:

• [x] convolutional_network.ipynb: tested in test_convnet.py
• [x] multilayer_perceptron.ipynb: untested
• [ ] softmax_regression.ipynb: untested
• [ ] dbm_demo/rbm.yaml: untested
• [x] deep_trainer/run_deep_trainer.py: tested in deep_trainer/test_deep_trainer.py
• [x] grbm_smd/cifar_grbm_smd.yaml: tested in grbm_smd/test_grbm_smd.py, see #67

For test_convnet.py, I copied and pasted the YAML code in convolutional_network.ipynb, changing a few things to make it run in a reasonable amount of time. It would be better to have the string be a template, and use Python formatting to fill in the value of hyper-parameters. If the .ipynb file can import that template string from an external file, that would be even better, since it would avoid duplicating it.

[memimo]

Perhaps we should make unit tests for scripts/papers too. Currently maxout/cifar10.yaml is broken.

[nouiz]

That is a good idea. But what is the run time of them?

On Tue, Jun 25, 2013 at 9:57 AM, Mehdi Mirza notifications@github.comwrote:

Perhaps we should make unit tests for scripts/papers too. Currently maxout/cifar10.yaml is broken.

— Reply to this email directly or view it on GitHubhttps://github.com/lisa-lab/pylearn2/issues/371#issuecomment-19977634 .