Causal Effect Inference for Structured Treatments
Authors: Jean Kaddour, Yuchen Zhu, Qi Liu, Matt J. Kusner, Ricardo Silva
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In experiments with small-world and molecular graphs we demonstrate that our approach outperforms prior work in CATE estimation. |
| Researcher Affiliation | Academia | Jean Kaddour Centre for Artificial Intelligence University College London Yuchen Zhu Centre for Artificial Intelligence University College London Qi Liu Department of Computer Science University of Oxford Matt J. Kusner Centre for Artificial Intelligence University College London Ricardo Silva Department of Statistical Science University College London |
| Pseudocode | Yes | b Pseudocode in a Py Torch-like style. |
| Open Source Code | Yes | Our PyTorch [43] implementation is online.2 2https://github.com/Jean Kaddour/SIN |
| Open Datasets | Yes | The Cancer Genomic Atlas (TCGA) simulation uses 9,659 gene expression measurements of cancer patients for covariates [62] and 10,000 sampled molecules from the QM9 dataset [46] as treatments. |
| Dataset Splits | No | Hyper-parameter tuning. To ensure a fair comparison, we perform hyper-parameter optimization with random search for all models on held-out data and select the best hyper-parameters over 10 runs. |
| Hardware Specification | No | The paper does not mention specific GPU, CPU, or other hardware models used for running experiments. It mentions "Azure cloud computing resources" but without specific hardware details. |
| Software Dependencies | No | Our PyTorch [43] implementation is online." and "We use the implementations of Py Torch Geometric [11]." The paper mentions PyTorch and PyTorch Geometric but does not provide specific version numbers for these libraries. |
| Experiment Setup | Yes | Input: Stage 1 data D1 := {(xi, yi)}m i=1, Stage 2 data D2 := {(xi, ti, yi)}n i=1 Step sizes λθ, λη, λψ, λφ. Number of update steps K. Mini-batch sizes B1, B2. |