#995 : documentation followup

This pull request introduces 2 alerts when merging b9eaf50 into a523de2 - view on LGTM.com

new alerts:

• 1 for Unreachable code
• 1 for Variable defined multiple times

This pull request introduces 2 alerts when merging 6ab3129 into 108af1a - view on LGTM.com

new alerts:

• 1 for Unreachable code
• 1 for Variable defined multiple times

Hi!
Will this new feature be released to a 1.0.4 or to a 1.1.0 version?
Will it be available soon?

Hi @CatOfTheCannals,
As you can tell by looking at the diff size, it's pretty big.
It will be released with Hydra 1.1, it will be available in master and later in dev or rc releases before Hydra 1.1 gets released.

@omry I don't know if I should open a feature request or if this is already possible, but suppose you have a complex hierarchy where most objects have to be recursively instantiated but for some we wish to defer the instantiation so that it can be managed inside the program. For example, if we wish to run large models sequentially on the same GPU, we wish to instantiate the second once we are done with the first. I had a feeling that _recursive_ special key could be used to achieve this so tried this configuration

bleurt:
    _recursive_: false
    _target_: datasets.load_metric

but I'm getting the error

Error in call to target 'datasets.load.load_metric':
TypeError("load_metric() missing 1 required positional argument: 'path'")
full_key: architecture.paraphrase_decoder.semantic_faithfulness_model

Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace```

which suggests that hydra does not care about _recursive_ . I definitely can't pass _recursive_ false to the the instantiation call for the entire object because the structure is complex and it would take a huge refactor to achieve this. Is this feature not supported?

@alexcoca I think you will need to pass _recursive_: false to the config that is one level above bleurt:

foo:
  _target_: bar
  _recursive_: false
  bleurt:
    _target_: datasets.load_metric

With this setup, I believe foo will be instantiated and bleurt will not.

@Jasha10 Yes, this was the only way to get things working. I removed bleurt and created a dummy function that returned datasets.load_metric as a work around as, bleurt was sadly at the same of the hierarchy with a bunch of objects that did need to be instantiated so I could not have _recursive_ at the level above. I'm happy enough with this workaround ... Having looked through the source code I understand that the logic is to avoid instantiating children of nodes where _recursive_: false is specified. Obviously, this would require that in my case the parent of bleurt specified _recusive_: false, as you suggest. Thanks for your help, I learned a lot.