Closed kasparvonbeelen closed 3 years ago
I hope this is done. I tried to replicate the code from Hu et al 2019 and fit it in the evaluation framework. Paper: Hu, R., Li, S., & Liang, S. (2019). Diachronic Sense Modeling with Deep Contextualized Word Embeddings: An Ecological View, 3899–3908. https://doi.org/10.18653/v1/p19-1379 -
Maybe @fedenanni can you have a look?
@kasra-hosseini if you have a bit of time in the next days, could you take a look at this? I am finishing the provenance PR with Mariona at the moment, otherwise max early next week I can go through this one
I think we need first to merge the provenance PR as this one includes some changes that are actually part of the provenance (the download scripts in utils). Just to avoid reviewing twice the same code
I think we need first to merge the provenance PR as this one includes some changes that are actually part of the provenance (the download scripts in utils). Just to avoid reviewing twice the same code
@fedenanni Yes, this builds on the output of the data produced by the Provenance PR (but there is no overlap I can think of, i.e. no download script in utils).
Sorry, I just meant that this PR would modify (by deleting) the download script in utils
It should be possible to run the scripts in #45 with the quotation_all_machine_nn01.pickle
file on SharePoint. I may have a made a mistake with linking the PR to branch, which 44-kNN-BERT-baseline
. But this can wait of course.
Sorry, I just meant that this PR would modify (by deleting) the download script in utils
Ah yes, sorry for that. I hope it won't delete the download file on dev
after merging? I just removed it, as it is not part of the current PR.
Eheheh don't worry! In the current setting it would, but we can rebase this branch right after we have merged the Provenance into Dev, so you will be completely aligned with dev apart from the new KNN changes - it's going to be fun!
Looking forward :-) .
@fedenanni Sorry for the delay. I will review this today.
@kasra-hosseini hey! don't worry at all!
Replicate method of (Wiedemann et al., 2019) and (Hu et al., 2019) as disambiguation baseline.