UDS-BOPE/
├── acquisition.py # Acquisition functions (Random, UCB, EI, LLM-EI)
├── model.py # Gaussian Process models (standard GP, preference GP)
├── preference_BO.py # Preference-guided BO for BH datasets
├── run_amide_coupling_bo.py # Amide coupling BO workflow
├── amide_llm_interface.py # LLM interface for amide reactions
└── LLM_survey/ # Survey generation and execution
├── run_llm_survey_BH.py # Buchwald-Hartwig survey
├── run_llm_survey_DA.py # Direct arylation survey
└── utils.py # LLM utilities
Buchwald-Hartwig datasets (BH1-5):
python preference_BO.py --n_trials 50 --n_start 1 --acc 0.7 --m 100 --beta 2.0Amide Coupling datasets:
python run_amide_coupling_bo.py --n_trials 50 --n_start 1 --acc 0.8 --m 100 --beta 2.0Arguments:
--n_trials: Number of independent BO runs (default: 50)--n_start: Number of initial random observations (default: 1)--acc: LLM survey accuracy for synthetic preference data (default: 0.7-0.8)--m: Number of pairwise comparisons (default: 100)--beta: Exploration parameter (not used in LLM-EI)
python LLM_survey/run_llm_survey_BH.py \
--questions questions.pkl \
--exp_database dataset.csv \
--res_dir results/ \
--llm_result survey_results.csv \
--model_type 0Model types: 0=Claude 3.5 Sonnet, 1=Claude 3 Sonnet, 2=Claude 3 Haiku
pip install -r requirements.txtTo check all installed package licenses:
pip install pip-licenses
pip-licenses --format=markdownFor detailed license information, see DEPENDENCIES.md.
See the LICENSE file for details.