Add ML for synthesis prototype to repository

[iburakov]

This PR's unfortunate fate is to become a huge solution hub for ML-related problems. Hope to merge it soon, so this is fixed. Contents:

• ML synthesis (Python codebase + Haskell integration)
• Dockerfile to build an environment for NITTA development / CI workflows
• ML synthesis CI workflow (crawl-train-test-release)

Resolves #235, #233, #231, #242

[ryukzak]

Very interesting py code style for me, I will steal some ideas. Also, now I understand why data scientists have not converted to programmers. Too much magic.

I commit some changes, which simplify notebook running automatically (not manually running of nitta, unix like paths, some warnings). Please, check it. I think it should continue work on windows, but not checked.

Did I correctly understand that example_model_1 is a model prepared by you manually?

What I think we need to do with this now:

• Experiment with nix as a repeatable cross-platform building tool  #171
• Add it to CI as a notebook with a smoke test and merge in master, because it will be dead very quickly without it. But how to organize a project correctly is an open question for me. Right now, I want to experiment with nix for that, but don't sure that is a good idea.
• Extract data retrieving parts to the library for profiling purposes.
• Starting the migration to a normal synthesis method is untimely and requires a lot of refactoring and some nitta side work, so we need to find someone to do this.

[iburakov]

@ryukzak Replying to your last message in this PR paragraph by paragraph.

The data science itself is pure magic, it's inevitable :)

I've looked at automations you made. It generally works on Windows too, but I'm not quite a fan of this experience, since something weird can happen. More on this later.

Yes, you've understood it right, example_model_1 is a manually prepared example model.

[iburakov]

When I was working on this prototype, I found that it's easier for debugging and playing to have a terminal near you and see what's exactly listening on a given port. Those automations may be necessary, but for a script (smoke testing script, for example). When it comes to playing with data in an interactive notebook, weird stuff happening in background can disorient. I stumbled upon hanging process, for example. The reasons are unknown :D

Note the mismatch between spi1 experiment and nodes count. It's the old process with example.lua is listening. Killing it helps:

ConnectionRefusedError happened sometimes too (delay of 0.5s is not enough in my case, perhaps). Reversing the lines (first doing read1() and then sleeping for a delay) helps. Did also some more tweaks in attempts to fix hanging process, but had no luck.

So in a playground it isn't worth it, probably :(

[iburakov]

Extract data retrieving parts to the library for profiling purposes.

I'll do that in a moment and push it here.

[iburakov]

Planning to add external synthesis script and only then request a review.