Code for Learning to Guide and to Be Guided in the Architect-Builder Problem

Requirements

• nop
• wandb (likely to have some login/setting up to do here)
• jax
• jaxlib
• pytorch
• readchar

Installation

This codebase uses aflred for experiment management. The code for the environment is hosted on a separated repo: https://github.com/flowersteam/architect-builder-env. To run experiments you need to make a new directory and clone the architect-builder-abig repo in it as well as the alfred and architect-builder-env repos.

mkdir abp
cd abp
git clone [email protected]:flowersteam/architect-builder-abig.git
git clone --branch abig-iclr2022 [email protected]:PBarde/alfred.git
git clone [email protected]:flowersteam/architect-builder-env.git

Then you need to install them using the following commands.

cd alfred
pip install -e .
cd ../architect-builder-env
pip install -e .
cd ../architect-builder-abig
pip install -e .

To make using alfred as seamless as possible, add the followings to your .bachrc:

alias alprep='python -m alfred.prepare_schedule'
alias allaunch='python -m alfred.launch_schedule'
alias alclean='python -m alfred.clean_interrupted'
alias alplot='python -m alfred.make_plot_arrays'
alias alretrain='python -m alfred.create_retrainbest'
alias albench='python -m alfred.benchmark'
alias alsync='python -m alfred.sync_wandb'
alias alcopy='python -m alfred.copy_config'
alias alupdate='python -m alfred.update_config_unique'

Running the experiments

All command lines should be ran from main_comem/main_comem. Each run takes approximately 48h to run on a CPU.

Reproducing results with 3 blocks

Prepare the experiments to be ran

alprep --desc 3b --schedule_file schedules/bw_task4_bc/grid_schedule_task_4.py

Run the experiments (recommended doing in parallel since each run takes 48h, by playing with allaunch arguments --n_processes and --n_experiments_per_proc)

allaunch --from_file schedules/bw_task4_bc/list_searches_bw_task4_bc.txt

Evaluate the performances

python utils/analyses_OOD.py --from_file schedules/bw_task4_bc/grid_schedule_task_4.py

Plot the results

python utils/make_analyse_OOD.plot --from_file schedules/bw_task4_bc/grid_schedule_task_4.py

Reproducing results with 6 blocks

Prepare the experiments to be ran

alprep --desc 6b --schedule_file schedules/bw_task4_bc_6b/grid_schedule_task_4.py

Run the experiments (recommended doing in parallel since each run takes 48h, by playing with allaunch arguments --n_processes and --n_experiments_per_proc)

allaunch --from_file schedules/bw_task4_bc_6b/list_searches_bw_task4_bc.txt

Evaluate the performances

python utils/analyses_OOD.py --from_file schedules/bw_task4_bc_6b/grid_schedule_task_4.py

Plot the results

python utils/make_analyse_OOD.plot --from_file schedules/bw_task4_bc_6b/grid_schedule_task_4.py

License

The project is licensed under the MIT license.

Citation

@inproceedings{
  barde2022learning,
  title={Learning to Guide and to be Guided in the Architect-Builder Problem},
  author={Paul Barde and Tristan Karch and Derek Nowrouzezahrai and Cl{\'e}ment Moulin-Frier and Christopher Pal and Pierre-Yves Oudeyer},
  booktitle={International Conference on Learning Representations},
  year={2022},
  url={https://openreview.net/forum?id=swiyAeGzFhQ}
}