A sequence-to-sequence model with attention and beam search for slot filling. Implemented in PyTorch and trained and evaluated on the ATIS dataset. Check the report for more implementation details.
- Clone the repository
- Install the required dependencies by running
pip install -r requirements.txt
- From the repository root, open and run the Jupyter notebook
Seq2Seq.ipynb
Note: the model was developed and tested using Python 3.7.10.
All the code to define the model and run the experiments is available in Seq2Seq.ipynb
.
The folder data contains the ATIS dataset, the conlleval.pl
script and the pre-trained embedding weights, obtained using the gensim
APIs and filtered to contain only the relevant tokens.
The project report is available in report.pdf
.