Robert Sachunsky

> The call was `python -m git_lfs -vv ~/git/$OPENSHIFT_APP_NAME.git $OPENSHIFT_REPO_DIR`. Oh, then my impression that this was also meant for **remote** repos (first cloning https or git URLs) was wrong...

Also, if I understand the [Seq2Seq implementation](https://github.com/farizrahman4u/seq2seq/blob/c020ccfc1fa3a651be272f8b4be48a10f9c3f0fa/seq2seq/models.py#L171) correctly, the peeky model does not drag the context vector along as additional, time-constant (repeated) input, like [2] and as described in the...

Thanks @MehmedGIT @joschrew for the explanation. Absolutely makes sense now – fantastic idea!

> An easy solution would be to ignore exists errors. Not sure though if this is good without the force flag? I agree – mkdir should be `mkdir -p` behaviour,...

> It looks like the CI tests are broken. Running less tests should not cause a failure of the remaining ones. It doesn't. Just the CI has not been run...

I suggest to implement [alignment path length as denominator](https://github.com/impactcentre/ocrevalUAtion/issues/21#issuecomment-839052773) instead of the GT length (which can be >1): https://github.com/qurator-spk/dinglehopper/blob/249787686f554ceee4a14c2610772095320d912a/qurator/dinglehopper/character_error_rate.py#L24-L28 (Ideally, you implement all 3 length options: alignment path, maximum sequence,...

Update: I recommend using rapidfuzz's `normalized_distance` instead of just dividing `distance` by the GT length. Internally (in the [CPP backend](https://github.com/rapidfuzz/rapidfuzz-cpp/blob/10426d24cd7479df0fe8c78b17877e756e1c3cd5/rapidfuzz/distance/Levenshtein_impl.hpp#L89-L99)) the denominator is calculated as the actual path length (=maximum...

@cneud your [requirement change in eynollah](https://github.com/qurator-spk/eynollah/commit/b3fa68439559479f2786c12482fd9270af9b4075) pinned that to 2.12. That in turn excludes using it together (in the same venv, as is most useful for ocrd_all) with sbb-binarization, which...

Or you wait for the PAGE-XML DOM to give us references upwards in the hierarchy. See OCR-D/core#313 and discussion there.

See https://github.com/apacha/sbb_binarization/pull/1 for an update of the PR.