Implementations for manuscript A model of conceptual bootstrapping in human cognition, by Bonan Zhao, Christopher G. Lucas, and Neil R Bramley.
causal_bootstrapping/
├── LICENSE
├── README.md
├── experiment
│ ├── css
│ │ └── style.css
│ ├── js
│ │ ├── config.js
│ │ ├── config_2.js
│ │ ├── config_3.js
│ │ ├── config_4.js
│ │ ├── config_5.js
│ │ ├── funcs.js
│ │ └── task.js
│ └── p
│ ├── task.html
│ └── welcome.html
├── models
│ ├── ag
│ │ ├── base_classes.py
│ │ ├── base_methods.py
│ │ ├── data
│ │ │ ├── task_frames.csv
│ │ │ ├── task_pm.csv
│ │ │ └── tasks
│ │ │ ├── exp_1.csv
│ │ │ ├── exp_2.csv
│ │ │ ├── exp_3.csv
│ │ │ └── exp_4.csv
│ │ ├── helpers.py
│ │ ├── program_inf.py
│ │ ├── program_lib.py
│ │ ├── program_sim.py
│ │ ├── sims
│ │ │ ├── process_combine.py
│ │ │ ├── process_construct.py
│ │ │ ├── process_decon.py
│ │ │ └── process_flip.py
│ │ └── task_terms.py
│ ├── gp
│ │ ├── gp_reg.py
│ │ └── gp_reg_results.csv
│ ├── pcfg
│ │ ├── Rational_rules.py
│ │ ├── prediction.py
│ │ └── training.py
│ └── requirements.txt
├── openai
│ ├── gpt3-predictions.csv
│ ├── gpt3-reports.csv
│ └── playground.ipynb
└── trials
├── data
│ ├── PM_LL.csv
│ ├── all_eqc.csv
│ ├── eig_trials.csv
│ ├── final_trials.csv
│ └── final_trials_2.csv
├── get_pms.py
├── get_trials.py
└── prep_pms.py
The experiment/ folder contains code for the online experiment. A live demonstration is at https://bramleylab.ppls.ed.ac.uk/experiments/bootstrapping/p/welcome.html
The models/ folder contains python code (python 3.9) for the adaptor grammar model (models/ag/), the rational rules model (models/pcfg/), and the gaussian process regression model (models/gp/).
For the adaptor grammar model, folder sims contains example scripts (process_construct.py, etc) to run the model. Modifications are needed to run the analyses in Supplementary Information. Contact me (Bonan Zhao, b.zhao@ed.ac.uk) for details.
In models/ag, there are object-oriented implementaion treating programs, terms and routers all as classes (base_classes.py, task_terms.py, program_lib.py). Correpondingly, their "causal influence" are implemented as functions and methods in the classes (base_methods.py, program_inf.py, program_sim.py and helpers.py). The data/ folder contains necessary prep data, eg. task setups.
The trials/ folder has python code for selecting generalization trials for Experiment 1 (see SI section in the linked manuscript).
Folder openai/ contains an ipython notebook that I used to batch-retrieve self-reports and generalization predictions from GPT-3, and the corresponding results (gpt3-reports.csv and gpt3-predictions.csv).
- Experiment data, pre-regs and data analysis scripts are available in OSF repo https://osf.io/9awhj/
- A live experiment demo: https://bramleylab.ppls.ed.ac.uk/experiments/bootstrapping/p/welcome.html
- Working github repo with all dev history and some un-reported attempts: https://github.com/zhaobn/comlog
- Cogsci paper Powering up causal generalization: A model of human conceptual bootstrapping with adaptor grammars