introduce hyper parameters and config

[elgehelge]

Part of #3393

[dmpetrov]

I did some basic testing. It looks really good! Thank you @elgehelge and @efiop!

Some issues that I was able to find:

1. Params list formatting should be human-readable like Parameter lr1, dropout are missing ...

$ dvc run -p lr1 -M metrics.json -f Dvcfile python train.py
Running command:
	python train.py
ERROR: failed to run command - Parameters '{'lr1'}' are missing from 'params.yaml'.

2. How to define nested params train.lr?

$ cat params.yaml
train:
    lr: 0.0041
    layers: 9
    dropout: 0.21

processing:
    th: 0.98
    bow: 7000

$ dvc run -p train.lr -M metrics.json -f Dvcfile python train.py
Running command:
	python train.py
ERROR: failed to run command - Parameters '{'train.lr'}' are missing from 'params.yaml'.

3. Can I redefine the params file name (or not in this change)?

[efiop]

Params list formatting should be human-readable like Parameter lr1, dropout are missing

Formatting issue, fixed.

How to define nested params train.lr?

Added support for basic a.b.c params support. Wildcards and more advanced things will be introduced later.

Can I redefine the params file name (or not in this change)?

Yes, -p path/to/params:param1,param2.

[efiop]

Thank you @elgehelge ! 🙏

[shcheklein]

@elgehelge @efiop do we have docs coming for this?

[efiop]

@shcheklein Not yet, this was just the first step. Docs will be later.

[jorgeorpinel]

Hi. So this is a hidden feature for now?

[efiop]

@jorgeorpinel Just unfinished and undocumented on dvc.org. Not hidden.