/black

/build

I tried to cover all the possible config cases in the unit tests.

Thanks.

/build

/build

I'm still looking into the existing implementations outside of monai...

for this pr,

• at the component config level looks like it is decoupled from the format-specific elements of json and yaml. could we have a yaml demo? looks like multiple research repositories are using yaml for workflow configs.
• the concept of unique id is unclear to me, why and where to ensure the uniqueness? also class_table[name] = modname doesn't allow the same class name under different packages, this could be an issue when importing from different packages
• the class scanner could be mostly replaced by pydoc.locate? see https://github.com/facebookresearch/hydra/blob/5f605a863dd4aed8f432884fcd1e1b388c8fb559/hydra/_internal/utils.py#L554-L558
  https://github.com/python/cpython/blob/main/Lib/pydoc.py#L1707
• this repo is achieving a similar goal https://github.com/catalyst-team/hydra-slayer
• the hydra implementation for class instantiation also looks powerful https://github.com/facebookresearch/hydra/blob/5f605a863dd4aed8f432884fcd1e1b388c8fb559/hydra/_internal/instantiate/_instantiate2.py#L121 (test cases: https://github.com/facebookresearch/hydra/blob/5f605a863dd4aed8f432884fcd1e1b388c8fb559/tests/instantiate/test_instantiate.py#L64), do we really want to implement from scratch here...

cc @ericspod @rijobro

Thanks for @wyli 's review and suggestions.

Hi @ericspod @rijobro ,

Could you please help also take a look and share some opinions?
I will come back to work on this ticket tomorrow.

Thanks in advance.

Hydra does do some of the things we want to include here, do we want to add it as a dependency? I'm not sure it could be an optional dependency since the model archiving functionality would need it to be there and it's a pretty important that it's present in MONAI.

this one also looks interesting https://github.com/jmespath/jmespath.py

this one also looks interesting https://github.com/jmespath/jmespath.py

Something structured like this looks like a good idea, I do like the "foo.baz" syntax. It's rather more complex that we need for this application perhaps, again though the concern is adding another dependency.

I have a bit of code in a tool for generating random data that instantiates objects from a YAML schema. This function to creates objects from the dictionary derived from the schema recursively so that new objects can be used as constructor arguments when instantiating others. The find_type_def is pretty well the same as get_class and much of the other code here is doing a very similar thing.

I'm good with this going forward if we want to keep the momentum going on the model archive.

Hi @ericspod @wyli ,

Thanks very much for your review and comments, suggestions.
I was busy with NVIDIA Clara issues when came back from the Chinese New Year holiday this week, now I am checking other repos' implementations of config parsing and trying to resolve your comments.

Thanks.

/black

/build

/build

/build

/black

/build

/build

/black

/build

I have merged this PR into the whole draft PR: #3593 and verified the overall features.
Will merge this PR and create following PRs for other parts.

Thanks.