- Tables of experiments implemented are as follows:
- The scripts can be run independently as they were developed on Google Colab first due to lack of access to a reliable GPU
- Due to lack of F1 score metric in Keras (https://github.com/keras-team/keras/wiki/Keras-2.0-release-notes#losses--metrics), I have created a F1 metric using Keras' Metric API for both F1-micro and F1-class for use in future work in punctuation prediction. They can be easily edited for use in other tasks as well.
- The MGB dataset (https://ieeexplore.ieee.org/document/7404863) was used for the experiments. The required files from this dataset are train.txt, dev.txt, train.ark, dev.ark, train.ctm and dev.ctm