returnn
returnn copied to clipboard
rwth-i6
Wei Zhou's solution for Pickle protocol incompatibility
During fast Baum-Welch training using the FST from Sprint, which currently relies on python2, there is a Pickle protocol incompatibility. This solution proposed by Wei Zhou solves this problem.
It seems this is heavily copied from ExternSprintDataset._read_next_raw (but code comments stripped away?), and now you would end up with three (almost? or even exact?) duplicate copies of the same code.
Why? We should not have duplicate code. You should introduce some small helper functions (or class) that you use in these three cases.
@Marvin84 or @ZhouW321 (or whoever else feels responsible for this, maybe @curufinwe ), can you clean this up and make ready for merge? We should not keep ugly custom patches around for too long. You should all be able to use the master as-is, and this should be useful for others (if you think what you do is useful for others).
this still needed, or outdated now?
I think it is neither. It might be still needed in some cases where you do some mixing of python versions. But I am not doing that so at least for my current setup this is not needed. I still feel that there might be some errors or inconsistencies in the current code, as there is insufficient test coverage for those parts. But I am not familiar enough with this code parts to really tell.
