Model-based Causal Bayesian Optimization
Authors: Scott Sussex, Anastasia Makarova, Andreas Krause
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically we find that MCBO compares favorably with existing state-of-the-art approaches. |
| Researcher Affiliation | Academia | Scott Sussex ETH Z urich scott.sussex@inf.ethz.ch Anastasiia Makarova ETH Z urich Andreas Krause ETH Z urich |
| Pseudocode | Yes | Algorithm 1 Model-based Causal BO (MCBO) and Algorithm 2 Model-based Causal BO with Hard Interventions (MCBO) |
| Open Source Code | Yes | We provide an open-source implementation of MCBO4. https://github.com/ssethz/mcbo |
| Open Datasets | Yes | We empirically evaluate MCBO on six problems taken from previous CBO or function network papers (Aglietti et al., 2020b; Astudillo & Frazier, 2021b). ... PSAGraph is inspired by the DAG from a real healthcare setting (Ferro et al., 2015)... |
| Dataset Splits | Yes | For GP-UCB and MCBO, β is chosen by cross-validation across tasks, as described in the appendix. ... The cross-validation for selecting β is performed across β = {0.05, 0.5, 5}. |
| Hardware Specification | No | The paper states 'EIFN and MCBO run on equivalent hardware, which has 4 times more ram than the hardware used for GPUCB and EICBO,' but does not specify exact models (e.g., CPU, GPU) or other detailed specifications. |
| Software Dependencies | No | The paper mentions software like 'Bo Torch package' and 'gradient-based optimizers' but does not provide specific version numbers for any libraries, frameworks, or operating systems. |
| Experiment Setup | Yes | The cross-validation for selecting β is performed across β = {0.05, 0.5, 5}. ... When parameterizing each ηi with a neural network, we always use a two layer feed-forward network with a Re Lu non-linearity, To map the output into [ 1, 1] we put the output of the network through an element-wise Sigmoid. ... In all noisy environments we estimate the expectation in the acquisition function (Eq. (12)) using a Monte Carlo estimate with 32 repeats for each gradient step. For noisy Dropwave we use 128 repeats because the environment is particularly noisy compared to other noisy environments. ... For MCBO and EIFN, we use identical values for all shared hyperparameters (e.g. GP kernels) and use the original hyperparameters from Astudillo & Frazier (2021b). |