[dev] Determine if namespace packaging will work for imports alongside pytorch.

[romesco]

Quoting @omry:

In an ideal world, the configs would be packaged with PyTorch itself, so probably:
torch.optim.AdamConf

I agree with this.

Alternatives are torch_config.optim.AdamConf. Using config.* is probably too generic of a module name and I imagine could lead to collisions.

[omry]

Keep in mind that whatever decision we have should either be friendly to what configen supports now or be added there.
I don't think we can do torch_config.optim.AdamConf right now.

see module_path_pattern.

[romesco]

Alright, until someone can raise an issue, we are going to stick with: hydra_configs.<module_path>. If others create their own configen configs such as: https://github.com/romesco/hydra-lightning for example, the idea is to ensure the package is set up as a compatible native namespace package. This means the module dir name is config and there is no __init__.py at the top level.

If those conventions are adhered to, many structured configs from different projects can be imported along side each other like in this example:

# structured config imports
from hydra_configs.torch.optim import AdamConf
from hydra_configs.torchvision.datasets import DatasetFolderConf
from hydar_configs.pytorch_lightning.trainer import TrainerConf

from
https://github.com/romesco/hydra-lightning/blob/50707694e7da1bab7753fb80402be7f26427f5f5/examples/mnist_00.py#L20-L22