Code for enumerative and neural synthesizers, with pragmatic listeners build on top.
To train a neural literal listener:
python train.py -h
Training programs used are in data/train_programs.json.
To run experiments and get results:
python experiment.py -h
Here, you can specify whether the listener type (neural/enumerative), listener variant (literal/pragmatic), and provide a path to one of the JSON files in specs which contain specifications. Results and logs are written to the output file.