fmelon/GeneratingSQLQueriesFromNaturalLanguage

Use of LSTMs and CNNs for generating SQL queries from natural language text

Generating SQL Queries from Natural Language

This repo provides extension to the implementation of SQLNet and added modules for using character embeddings along with word embeddings for training the bidirectional LSTMs.

The code is implemented in python2.7 using the PyTorch 0.3.1. For install other requirements execute requirements.txt as

pip install -r requirements.txt

We have installed and extracted the pretrained GLOVE embeddings which can be used directly for training and testing the model. We have added a class inside word_embedding.py called CharacterEmbedding which contains modules and components related to character embeddings.

For training the model, we need to run "train.py" with the following commands:

For training Seq2SQL use

python train.py --baseline

For training SQLNet with column attention use (1)

python train.py --ca

For training SQLNet using trainable embeddings first execute the previous command(1) then execute

python train.py --ca --train_emb

For training SQLNet using character and word embeddings(our implementation) use

python train_embed_char.py --ca

For testing the model first execute any of the above train commands which trains and saves the best model, then execute the corresponding test command as follows:

For testing Seq2SQL use

python test.py --baseline

For testing SQLNet with column attention use (1)

python test.py --ca

For testing SQLNet using trainable embeddings first execute the previous command(1) then execute

python test.py --ca --train_emb

For testing SQLNet using character and word embeddings(our implementation) use

python test_embed_char.py --ca