1anw - add HPS tutorial on image classification, CORDS, ResNet architecture

[1anw]

The file should be found at tutorials/notebooks/09_HPS_CORDS. I also included a way to use ConfigSpace with a real hyperparameter space.

[Deathn0t]

Hello @1anw,

Thank your contributing this tutorial. The proposed notebook does not contain outputs which are required to be presented on the deephyper.readthedocs.io website.

[Deathn0t]

Hello @1anw and @BethanyL , I am adding Bethany to the thread because I think Ian ended is internship already. I took an other look at the new commits, the tutorial looks interesting and quite well shaped but I think some editions are still necessary to publish it. Let me explain:

In this one PR I am finding 4 "examples" about how to use CORDS:

• notebooks/09_HPS_CORDS/tutorial_09.ipynb: this notebook does not have any pre-computed outputs which makes it not publishable, because outputs are used to build the final HTML published on the website (and to make it readable without requiring execution).
• scripts/04_CORDS_Training/*: here there are 3 files. The rst file (the one published on the website does not follow the two scripts, it uses the thread evaluator when the two scripts are using the mpicomm evaluator. I propose to focus on using the mpicomm evaluator, keep only one script or clarify the difference between cifar10tutorial.py and cifar10cordsmodel.py (you can have subsections in the rst to detail each one.

I also propose to remove unused imports in scripts/tutorials to clarify the code.

If you could help me understand the idea behind all these files it would be great. I think it is better to have 1 page about CORDS probably starting with serial execution (default evaluator) then adding details about how to adapt in a supercomputing case (from 1 process, to multiple CPU processes, then GPU processes).

Do not hesitate to explain me if I missed something.
Best regards,
Romain.