A semantic parser implemented using a sequence-to-sequence neural network model with attention. Much of this code is based on the implementation within the Tensorflow Codebase.
- Tensorflow version 0.9.0 (Other versions may work, but this is the only one that has been tested)
- NLTK
The parser is run using the following command:
python run_parser.py {train, test} [optional arguments]
A full description of required and optional arguments can be found by running python run_parser.py -h
Of particular note is the option -d, which lets you specify the directory where the training/testing data is located.
Within the specified directory, the system will look for files called train.txt and test.txt containing train and test data, respectively.
The project is organized into the following files/directories:
run_parser.py: Contains the code to run parser training/testing.parser_model.py: Contains the implementation of the parser models.parse_input.py: A script that runs the parser on individual inputs.config.py: Contains a class that holds parser configuration and hyperparameter settings.data_utils.py: Contains methods assisting in the loading/processing of training and testing data.data/: Contains datasets for experimentation. The following datasets are included: