Code repository of "LancBiO: Dynamic Lanczos-aided Bilevel Optimization via Krylov Subspace".
- Ubuntu 20.04
- Python 3.8
- PyTorch 1.13.1
- CUDA 11.7
- wandb 0.16.0 (We heavily rely on Weights & Biases for visualization and monitoring)
You can create a conda environment by simply running the following commands.
$ conda create -n lancbio_env python=3.8
$ pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117
$ pip install wandbTo start training, run
$ conda activate lancbio_env
$ cd ./Code/LancBiO_deterministic
$ python3 main.py$ conda activate lancbio_env
$ cd ./Code/LancBiO_stochastic
$ python3 main.py$ conda activate lancbio_env
$ cd ./Code/SytheticProblem
$ python3 main.py