_inference within create_supervised_trainer?

Hi,

In one of ignite’s issues (https://github.com/pytorch/ignite/pull/48/files) I saw that the _inference was passed in to the trainer engine. Why is this option not in the docs? Is it not recommended to do this? Why not?

Thanks

Moving to Ignite category for better visibility and CC’ing @vfdev-5 :slight_smile:

2 Likes

Seems like it is a very old PR and things changed since than.

Currently, we do not provide such abstractions like Trainer etc as there is no unique definition of it as it can be task/config dependent etc. We left that to the user to compose its trainer using one or more Engines for training/validation etc.

1 Like