Natural-Language-To-Bash-Commands-Translation

This project is aimed at exploring different methods to convert a given natural language description (a given requirement in natural language) to a single line bash command.

Dependencies

1. pytorch==1.7.1
2. bashlint
3. spacy

Data

The train/valid/test splits of the dataset can be found in the 'data' folder.

Data Split	Size
train	25159
dev	5159
test	5159

Models

1. Seq2Seq uses a simple RNN based model to implement the above problem.
2. Seq2Seq_Attention uses Bahdanau attention along with the RNN based model to achieve better performance.
3. Seq2Seq_Attention_ut has an added auxiliary part to predict the utilities which should be present in a command and help the decode produce better results. The model is explained through a disgram in 'model.pdf'.
4. Seq2Seq_Transformers has a transformer encoder decoder architecture to achieve the above job. It gives the best result among all the methods.

References

1. Nl2Cmd Competition
2. Project CLAI