Extrinsic task code

[stephantul]

Hi,

I was wondering where I could find the testing/training code for the extrinsic tasks, i.e., SST-2 and CoNLL-03.
Also, are the models included in the repository the models for which you report the scores in the paper?

Thanks!

[tigerchen52]

Hi,

I was wondering where I could find the testing/training code for the extrinsic tasks, i.e., SST-2 and CoNLL-03. Also, are the models included in the repository the models for which you report the scores in the paper?

Thanks!

Hi,

Thanks for your interest in our work!

As for the extrinsic evaluations, I have added some files for this goal.
You can follow the instructions in the README to reproduce the reported scores.

1. SST-2: cnn_text_classification
2. CoNLL-03: rnn ner

Besides, we provided two model files, love_fasttext and love_bert_base_uncased.
Note that the love_fasttext model is the same as the file in output/model_ merge.pt.

Best,
Lihu

[stephantul]

Hi,

Thanks for the quick response. I'll check out the code! The model namespace mentions that it was trained using data/wiki_100.vec. But is this correct? I'm assuming that this is a placeholder file, rather than the real data the model was trained on.

Thanks!

[tigerchen52]

Hi,

Thanks for the quick response. I'll check out the code! The model namespace mentions that it was trained using data/wiki_100.vec. But is this correct? I'm assuming that this is a placeholder file, rather than the real data the model was trained on.

Thanks!

Yes, your understanding is totally correct.
data/wiki_100.vec is just an example to test the code. You can download your personal target pre-trained word embeddings, e.g., Fasttext.
As mentioned before, LOVE can mimic various existing word embeddings.

[pestrstr]

Hi!
Are you also planning to add the code for reproducing results for extrinsic tasks on MR and BC2GM datasets?
Thanks!

[tigerchen52]

Hi! Are you also planning to add the code for reproducing results for extrinsic tasks on MR and BC2GM datasets? Thanks!

Hi, for the two datasets, the only thing to do is to put the files into the corresponding task path.
For example, MR is a text classification dataset, just replace the file in cnn_text_classification.
Same case for the BC2GM. Note that the file format should also be the same.

[pestrstr]

Thank you for your answer!