neural_qa

Source code for "Neural Multi-Step Reasoning for Question Answering on Semi-Structured Tables"

The code base is slightly convoluted, if you have any questions do not hesitate to contact: [email protected]

Generate Training and Test Data

• Get code from Compositional Semantic Parsing on Semi-Structured Tables (ACL 2015) (https://worksheets.codalab.org/worksheets/0xf26cd79d4d734287868923ad1067cf4c/)
• Overwrite code from 1) with code from src_neural
• Run code according to instructions from 1), during run files starting with "translated_" will be created, they will contain the training and test data. Note: Scala is a requirement

Create Vocabulary

• Run ./vocab_creator.py to create the vocabulary based on the training and test data (needs to be given as one file as input).

Training / Testing

See main.py. Note: for testing make sure to train a model first and specifiy the persisted model in the parameters.