Multi-Fidelity Active Learning with GFlowNets

This repository extends the code used in the paper Multi-Fidelity Active Learning with GFLowNets, implemented in github.com/nikita-0209/mf-al-gfn.

Installation

• This code requires python 3.10.
• If you are installing this in a compute cluster (for example, Mila's), you can load the required modules by running source ./prereq_cluster_cpu.sh.
• Setup is currently only supported on Ubuntu. It should also work on OSX, but you will need to handle the package dependencies.
• The recommended installation is as follows:

python3.10 -m venv ~/venvs/activelearning  # Initalize your virtual env.
source ~/envs/activelearning/bin/activate  # Activate your environment.
./prereq_python.sh  # Updates pip and forces the installation of potentially problematic libraries
python -m pip install .[all]  # Install the remaining dependencies of this package.

The above steps install PyTorch for CPU only. In order to install a cuda-enabled PyTorch, it must use the wheels for cuda 11.8 in order to be compatible with the GFlowNet package. In the cluster, first run source ./prereq_cluster_gpu.sh and then install PyTorch:

python -m pip install torch==2.0.1 --index-url https://download.pytorch.org/whl/cu118

Use the following commands to install FAENet:

pip install git+https://github.com/RolnickLab/ocp.git@uncertainty

For development you can use a local installation of the package:

git clone https://github.com/RolnickLab/ocp.git
cd ocp
pip install -e .

Run Examples

To run the different examples you can use the following command:

python main.py --config-name <config-filename> user=<user-filename>

Pre-Defined Config Files

The following pre-defined config files are available:

• test_branin
• test_hartmann
• test_ocp

Customize Config Options

Some config options can be customized independently of the task. Here is a list of possible options for each component:

• sampler: random | greedy | random_gflownet | gflownet
• selector: selector | score
• surrogate: gp | dkl | svdkl_kernel_wrapper
• acquisition: botorch_ei | botorch_mve | botorch_nei
• user: default | <custom_user_file>
• logger: wandb | base

Citation

If you use this code for your own work, please consider citing our published work:

@misc{hernandezgarcia2023multifidelity,
      title={Multi-Fidelity Active Learning with GFlowNets}, 
      author={Alex Hernandez-Garcia and Nikita Saxena and Moksh Jain and Cheng-Hao Liu and Yoshua Bengio},
      year={2023},
      eprint={2306.11715},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}