Update Lasagne installation doc to new gpuarray backend

[dreamflasher]

The documentation at http://lasagne.readthedocs.io/en/latest/user/installation.html to verify cuda and cudnn installation is outdated. It currently yields: ValueError: You are tring to use the old GPU back-end. It was removed from Theano. Use device=cuda* now. See https://github.com/Theano/Theano/wiki/Converting-to-the-new-gpu-back-end%28gpuarray%29 for more information.
Unfortunately the documentation at Theano does not contain the information to verify cuda and cudnn installation.

[f0k]

Yes, this documentation has not been updated for Theano 0.10+ yet, it will be updated at the latest when Lasagne drops support for Theano 0.8 (the last version of Theano that did not deprecate the old backend).
This being said, I can't find an equivalent for

THEANO_FLAGS=device=gpu python -c "import theano; print(theano.sandbox.cuda.device_properties(0))"

in the new backend. To verify whether cuDNN is available under the new backend, you can run:

python -c "from theano.gpuarray.dnn import dnn_present as d; print(d() or d.msg)"

[nouiz]

If you use device=cuda for the new back-end, you have this printed during the import: Using cuDNN version 7001 on context None Mapped name None to device cuda: Tesla P100-PCIE-12GB (0000:03:00.0) So you see which version of cudnn is being used. It also show you the GPU used (name and PCI-E bus in case you have multiple GPU in the same computer) The installation is partialy tested during import. So you are sure we can use cudnn and most of cuda. We do not test all libraries at import. We didn't reimplement device_properties in the new back-end.

…

On Fri, Feb 16, 2018 at 9:11 AM Marcel Ackermann ***@***.***> wrote: The documentation at http://lasagne.readthedocs.io/en/latest/user/installation.html to verify cuda and cudnn installation is outdated. It currently yields: ValueError: You are tring to use the old GPU back-end. It was removed from Theano. Use device=cuda* now. See https://github.com/Theano/Theano/wiki/Converting-to-the-new-gpu-back-end%28gpuarray%29 for more information. Unfortunately the documentation at Theano does not contain the information to verify cuda and cudnn installation. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#897>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALC-7uO3mMd8Qc7zDk9o3GCR-qcMO38ks5tVYyvgaJpZM4SIYru> .

[f0k]

If you use device=cuda for the new back-end, you have this printed during the import:

True, there's no need to query the device properties, simply importing Theano is enough to see whether the CUDA and cuDNN are used. The second command can still be useful for troubleshooting in case cuDNN is not being used.

I'm leaving this open since the documentation still needs to be updated later. (This requires more changes, it should include a section on libgpuarray/pygpu.)