Comparing LSTM and its variants for Parts of Speech Tagging problem

Mini project submitted as part of M.Tech course in semester 2.

Setup

This project is based on Python 3 and requires NVIDIA-GPU with CUDA 9.0 to run PyTorch with GPU computing capabilities.

Dataset

Penn Tree bank and Brown datasets were used for the experiments.

Details

documents folder contains the base research paper(which was referred for the project) and the project report.

LSTM variants used(from the paper):

• Standard LSTM Cell
• LSTM Cell without the forget gate
• LSTM Cell without the output gate
• LSTM Cell without the input gate
• LSTM cell with forget gate bias set to 1

Code files

• main.py contains the script for the experiments(training and validation)
• TreeBankDataSet.py contains data pre-processing steps
• pos_tagger.py contains the architecture of generic pos_tagger
• lstm.py contains variants of LSTM Cell mentioned above.