CUNY-CL / udtube

Neural morphological analyzer
Apache License 2.0
4 stars 3 forks source link

Better exposes important callbacks #50

Closed kylebgorman closed 1 month ago

kylebgorman commented 1 month ago

This is my attempt to address #41. It creates new subconfigs" (cf. trainer, model, data) for the prediction callback and for model checkpointing. As part of this process I also made it so that if no path for the output predictions is specified, it is just written to stdout; thus this indirectly addresses #45 (and I have merged in #49 to help).

Ideally I would have liked to be able to force users to specify a prediction path or model checkpointing, but I couldn't figure out how, and this is the next best thing (and is arguably more LightingCLI-like): it makes them much more transparent and substantially shortens the amount of boilerplate necessary to specify the.

Closes #41.

DanielYakubov commented 1 month ago

can you merge the conllu PR to main before this and I'll review again? Having trouble understanding what's what.

kylebgorman commented 1 month ago

can you merge the conllu PR to main before this and I'll review again? Having trouble understanding what's what.

Fair, and will do.

kylebgorman commented 1 month ago

can you merge the conllu PR to main before this and I'll review again? Having trouble understanding what's what.

Fair, and will do.

Done. Best way to understand this is to look how I changed the .yaml files in config.