giancarlok / snli-entailment Goto Github PK

Implementation of a attention model for entailment (for fun with keras) from this paper.

UPDATE: The code has been migrated to functional api. Works with Keras 1.0.6

The code is research quality -- which means there are no guarantees!

It works, but somewhat poorly than the numbers in the paper.

To train,

• Download snli dataset.
• Create train, dev, test files with tab separated text, hypothesis and label (example file train10.txt). You can find some snippet in reader.py for this, if you are lazy.
• Train!

You should be able to get >70% test and dev accuracy (I did no tuning).

Log is written out in *.log file with callback for accuracy.

For comments, improvements, bug-reports and suggestions for tuning, email [email protected]