Right now the logic about setting up the pre-trained Hugging Face encoders lives in a protected method of the UDTube module. This probably unnecessarily extends this module, and there is basically a if/elif/else cascade to deal with special cases that's likely to get longer. Here is what I'd suggest as an alternative:
Create encoders.py and move the special casing code there.
Sort things into three (notional) lists:
Encoders we've tried (possibly with special case) and know to work; use the special case and return the encoder
Optionally, encoders we know don't work: raise an module-specific exception
For all other encoders, just look it up on Hugging Face and see if it works with we think the defaults are; try it and return it
Then, we just call into this module in the constructor for UDTube.
This is slightly cleaner and stores code with lots of special casing in a module that most users will never have to look at.
Right now the logic about setting up the pre-trained Hugging Face encoders lives in a protected method of the
UDTubemodule. This probably unnecessarily extends this module, and there is basically a if/elif/else cascade to deal with special cases that's likely to get longer. Here is what I'd suggest as an alternative:encoders.pyand move the special casing code there.Then, we just call into this module in the constructor for
UDTube.This is slightly cleaner and stores code with lots of special casing in a module that most users will never have to look at.