Flair: Introducing Attentional BiLSTM for Text Classification

[53X]

As per my understanding , Flair by default uses the standard BiLSTM model for text classification.

I guess introducing Attentional BiLSTM might improve this maybe?

An easy solution to this might be to just introduce a parameter named attention in the TextClassifier class that decides whether to use attention or not

[53X]

If this issue is a go, I would love to take up the challenge

[alanakbik]

That's a great idea - we'd very much appreciate it if you added this! :)

[53X]

@alanakbik , I am on it

[nickkimer]

@53X excited to see this implementation

[heukirne]

Here is an interesting approach implemented also in PyTorch that could be useful:
https://github.com/hantek/SelfAttentiveSentEmbed#third-party-implementations

[53X]

@heukirne , we could introduce this as well as another improved version of the Text Classifier.
BTW I will make this PR in another 2 days.... I was out :)

[53X]

@alanakbik, can this be merged if this is working fine now

[alanakbik]

Ok, I've just put in a PR #582 to merge the branch into master. Could you double-check that everything looks OK from your side in the PR?

[53X]

this looks ok @alanakbik

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[HiyaToki]

Was this feature removed from later versions?