Counterfactual Prediction for Outcome-Oriented Treatments
Authors: Hao Zou, Bo Li, Jiangang Han, Shuiping Chen, Xuetao Ding, Peng Cui
ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results on both synthetic datasets and semi-synthetic datasets demonstrate the effectiveness of our method. |
| Researcher Affiliation | Collaboration | 1Department of Computer Science and Technology, Tsinghua University, Beijing, China; email:zouh18@mails.tsinghua.edu.cn, cuip@tsinghua.edu.cn 2School of Economics and Management, Tsinghua University, Beijing, China; email: libo@sem.tsinghua.edu.cn 3Meituan, Beijing, China; {hanjiangang,chenshuiping,dingxuetao}@meituan.com. |
| Pseudocode | Yes | Algorithm 1 Outcome-Oriented Sample Re-weighting(OOSR) |
| Open Source Code | No | The paper does not provide any statement or link indicating the availability of open-source code for the described methodology. |
| Open Datasets | Yes | The confounder feature is obtained from a real-world dataset TCGA (Weinstein et al., 2013). |
| Dataset Splits | No | A sample set of size 10000 is randomly generated as held-out test-set to compute the out-of-sample metric. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper mentions software components like 'neural networks', 'ELU activation function', and 'SGD optimizer', but does not provide specific software dependencies with version numbers (e.g., PyTorch 1.9, TensorFlow 2.x). |
| Experiment Setup | Yes | The predictive model is a neural networks with two hidden layers of size 20. The policy networks in IPS-Bandit Net is of the same architecture. We use the ELU activation function. The predictive models are trained by SGD optimizer for 60000 iterations in synthetic experiments, 100000 iterations in setting 1 of semi-synthetic experiments and 300000 iterations in setting 2 of semi-synthetic experiments. The policy networks are trained for 4000 epochs. For our algorithm, in each experiment, the length of the first stage is 40% of the training process, and the length of each round in the second stage is 5% of the training process. We choose "lambda" = 10.0 and "tau" = 0.2. |