AMAN SINGH
python entailment_recognition.py
open the .ipynb
file under exploration folder in jupyter notebook
I was able to get the model to an accuracy of about 72.48% using a keras sequential model with a GRU layer. Simplifiying the model made it severly underfit, adding more complexitiy to the architecture of the model made it start overfitting after jsut 4-5 epochs. This is the best I could get to in this time period.
This has been one of the many interesting problem statements i have had recently. Also I had to look about & understand what exactly is meant by entailments
. It was a good learning. Going forward I'd like to try out ensemble technique with the basic ML classifiers and also try out experimenting with other architecture of Neural Nets. The neural net architecture I have used is quite similar (with some differneces ofcourse) to any neural net that would be used for basic classification tasks like sentiment analysis, etc.
I tried a CNN architecture with just 1 layer of convolution & 1 max pooling, going forward I'd like to experiment with some tweaks on that architecture as well.