lightbot-grammar-induction

Analysis code for "Exploring the hierarchical structure of human plans via program generation"

Code structure

Generating figures from the paper

• journal/examples.ipynb - Generates pictures of the task, along with qualitative examples.
• journal/predict-programs.ipynb - Program enumeration and model fitting.
• journal/descriptive.ipynb - A number of more descriptive analyses.

File listing

Here's a partial file listing, noting the files that contain the most important elements of the model.

• experiment/data/human_raw/0.4/*.csv has raw data from the experiment.
• data/programs.csv is the data in a format that is easier to work with, focusing on the data analyzed in the paper.
• lb/search.py has A* search.
• lb/heuristics.py has the heuristic based on an approach for the traveling salesman problem.
• lb/program_generation.py the greedy algorithm for generating programs.
• lb/scoring.py computes the various priors tested in the paper.
• lb/fitting.py has model fitting code, as well as code orchestrates the entire analysis (it runs trace search, program generation, etc).
• lb/simple-ag.py has a simplified reimplementation of the grammar induction prior that is compared to the results of sampling from a generative algorithm for the prior. Not used in analysis, but may be helpful for understanding the model.

Install dependencies

pip install -r requirements.txt

Run tests

make test