TaicaiChen/PG-LBO

[AAAI2024]PG-LBO: Enhancing High-Dimensional Bayesian Optimization with Pseudo-Label and Gaussian Process Guidance

PG-LBO

This is the official repository for our AAAI 2024 paper:

PG-LBO: Enhancing High-Dimensional Bayesian Optimization with Pseudo-Label and Gaussian Process Guidance

Environment installation

• Install the experiments running environment.

conda env create -f environment.yml
conda activate pglbo_env

• Set result storage path.

echo '~/LSO-storage/' > ./utils/storage_root_path.txt

Usage guide

Topology task

Set up data

• The data can be downloaded from the following git repo: Neural Networks for Topology Optimization then place the unzipped files into your DATA_STORAGE_ROOT/data/topology_data/ (see storage_root_path.txt to check your DATA_STORAGE_ROOT).

Run the experiment.

• Train VAE.

chmod u+x ./weighted_retraining/scripts/models/train-topology-pseudo.sh
./weighted_retraining/scripts/models/train-topology-pseudo.sh

• Execute the optimization process.
  • Set parameters for pseudo-label training and GP-guided techniques.

  ....
  
  #-- Choose whether to use pseudo-label training --#
  use_pseudo='--use_pseudo' #'' represents not use.
  pseudo_data_size=5000
  pseudo_beta=0.5
  beta_pseudo_loss=0.5
  beta_pseudo_loss_final=0.75
  
  #-- Choose the pseudo data sampling you want to use --#
  pseudo_sampling_type_ind=4
  pseudo_sampling_type=("" "random" "cmaes" "ga" "noise")
  pseudo_sampling_type_kws=("" "{}" "{'n_best':100,'n_rand':0,'sigma':0.25,'use_bo':False}" "{'n_best':100,'n_rand':0,'use_bo':False}" "{'n_best':100,'n_rand':0,'sigma':0.1,'use_bo':False}")
  if ((pseudo_sampling_type_ind == 0)); then pseudo_sampling=''; else pseudo_sampling="--pseudo_sampling_type ${pseudo_sampling_type[$pseudo_sampling_type_ind]}"; fi
  if ((pseudo_sampling_type_ind == 0)); then pseudo_sampling_kw=''; else pseudo_sampling_kw="--pseudo_sampling_type_kw ${pseudo_sampling_type_kws[$pseudo_sampling_type_ind]}"; fi
  
  #-- Choose whether to use GP guidance --#
  use_ssdkl='--use_ssdkl' # '' represents not use.
  beta_ssdkl_loss=1 # weight of GP guidance loss.
  
  ....

  • Execute the script.

  chmod u+x ./weighted_retraining/scripts/robust_opt/robust_opt_topology_pseudo.sh
  ./weighted_retraining/scripts/robust_opt/robust_opt_topology_pseudo.sh

Expression task

Set up data

• Set up the data with expr_dataset.py after downloading and unzipping expression data from weighted-retraining repo

# download necessary files and unzip data
url=https://github.com/cambridge-mlg/weighted-retraining/raw/master/assets/data/expr/
expr_dir="./weighted_retraining/assets/data/expr"
mkdir $expr_dir
for file in eq2_grammar_dataset.zip equation2_15_dataset.txt scores_all.npz;
do
 cmd="wget -P ${expr_dir} ${url}${file}"
 echo $cmd 
 $cmd
done
unzip "$expr_dir/eq2_grammar_dataset.zip" -d "$expr_dir"

# split data and generate datasets used in our BO experiments
python ./weighted_retraining/weighted_retraining/expr/expr_dataset.py \
           --ignore_percentile 65 --good_percentile 5 \
           --seed 0 --save_dir weighted_retraining/data/expr

Run the experiment.

• Train VAE.

chmod u+x ./weighted_retraining/scripts/models/train-expr-pseudo.sh
./weighted_retraining/scripts/models/train-expr-pseudo.sh

• Execute the optimization process.
  • Set parameters for pseudo-label training and GP-guided techniques.

  ....
  
  #-- Choose whether to use pseudo-label training --#
  use_pseudo='--use_pseudo' #'' represents not use.
  pseudo_data_size=20000
  pseudo_beta=0.5
  beta_pseudo_loss=0.1
  beta_pseudo_loss_final=0.75
  
  #-- Choose the pseudo data sampling you want to use --#
  pseudo_sampling_type_ind=4
  pseudo_sampling_type=("" "random" "cmaes" "ga" "noise")
  pseudo_sampling_type_kws=("" "{}" "{'n_best':100,'n_rand':0,'sigma':0.25,'use_bo':False}" "{'n_best':100,'n_rand':0,'use_bo':False}" "{'n_best':100,'n_rand':0,'sigma':0.1,'use_bo':False}")
  if ((pseudo_sampling_type_ind == 0)); then pseudo_sampling=''; else pseudo_sampling="--pseudo_sampling_type ${pseudo_sampling_type[$pseudo_sampling_type_ind]}"; fi
  if ((pseudo_sampling_type_ind == 0)); then pseudo_sampling_kw=''; else pseudo_sampling_kw="--pseudo_sampling_type_kw ${pseudo_sampling_type_kws[$pseudo_sampling_type_ind]}"; fi
  
  #-- Choose whether to use GP guidance --#
  use_ssdkl='--use_ssdkl' # '' represents not use.
  beta_ssdkl_loss=0.1 # weight of GP guidance loss.
  
  ....

  • Execute the script.

  chmod u+x ./weighted_retraining/scripts/robust_opt/robust_opt_expr_pseudo.sh
  ./weighted_retraining/scripts/robust_opt/robust_opt_expr_pseudo.sh

Molecule task

Set up data.

• To download and build the Zinc250k dataset with black-box functions labels, execute setup-chem.sh

# download necessary files
url=https://github.com/cambridge-mlg/weighted-retraining/raw/master/assets/data/chem_orig_model
molecule_dir="./weighted_retraining/assets/data/chem_orig_model/"
mkdir $molecule_dir
for file in train.txt val.txt vocab.txt README.md;
do
 wget -P $molecule_dir "$url/$file"
done

# preprocess molecule data for BO experiments
chmod u+x ./weighted_retraining/scripts/data/setup-chem.sh
./weighted_retraining/scripts/data/setup-chem.sh

Run the experiment.

• get pre-train model

url=https://github.com/cambridge-mlg/weighted-retraining/raw/master/assets/pretrained_models/chem.ckpt
wget -P ./weighted_retraining/assets/pretrained_models/chem_vanilla/ $url

• Execute the optimization process.
  • Set parameters for pseudo-label training and GP-guided techniques.

  ....
  
  #-- Choose whether to use pseudo-label training --#
  use_pseudo='--use_pseudo' #'' represents not use.
  pseudo_data_size=100000
  pseudo_beta=0.5
  beta_pseudo_loss=0.1
  beta_pseudo_loss_final=0.75
  pseudo_train_data_path="weighted_retraining/data/chem/zinc/orig_model/tensors_train_pseudo"
  pseudo_val_data_path="weighted_retraining/data/chem/zinc/orig_model/tensors_val_pseudo"
  
  #-- Choose the pseudo data sampling you want to use --#
  pseudo_sampling_type_ind=4
  pseudo_sampling_type=("" "random" "cmaes" "ga" "noise")
  pseudo_sampling_type_kws=("" "{}" "{'n_best':100,'n_rand':0,'sigma':0.25,'use_bo':False}" "{'n_best':100,'n_rand':0,'use_bo':False}" "{'n_best':100,'n_rand':0,'sigma':0.1,'use_bo':False}")
  if ((pseudo_sampling_type_ind == 0)); then pseudo_sampling=''; else pseudo_sampling="--pseudo_sampling_type ${pseudo_sampling_type[$pseudo_sampling_type_ind]}"; fi
  if ((pseudo_sampling_type_ind == 0)); then pseudo_sampling_kw=''; else pseudo_sampling_kw="--pseudo_sampling_type_kw ${pseudo_sampling_type_kws[$pseudo_sampling_type_ind]}"; fi
  
  #-- Choose whether to use GP guidance --#
  use_ssdkl='--use_ssdkl' # '' represents not use.
  beta_ssdkl_loss=0.1 # weight of GP guidance loss.
  
  ....

  • Execute the script.

  chmod u+x ./weighted_retraining/scripts/robust_opt/robust_opt_chem_pseudo.sh
  ./weighted_retraining/scripts/robust_opt/robust_opt_chem_pseudo.sh

Citation

@misc{chen2023pglbo,
      title={PG-LBO: Enhancing High-Dimensional Bayesian Optimization with Pseudo-Label and Gaussian Process Guidance}, 
      author={Taicai Chen and Yue Duan and Dong Li and Lei Qi and Yinghuan Shi and Yang Gao},
      year={2023},
      eprint={2312.16983},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Acknowledgements

• Thanks to the authors of the High-Dimensional Bayesian Optimisation withVariational Autoencoders and Deep Metric Learning for providing their implementations of high-dimensional BO based on weighted-retraining VAE, which we based most of our code on.