Inferring the Long-Term Causal Effects of Long-Term Treatments from Short-Term Experiments
Authors: Allen Tran, Aurelien Bibaut, Nathan Kallus
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Section 4 uses simulated data to evaluate our method against a range of alternatives, as well as exploring robustness to real world complications. |
| Researcher Affiliation | Collaboration | 1Netflix Inc., Los Angeles, USA 2Netflix Inc., Los Gatos, USA 3Cornell University, New York, USA. |
| Pseudocode | No | The paper does not contain any pseudocode or algorithm blocks. |
| Open Source Code | Yes | 4Code and data for the experiments is available at: https://github.com/allentran/long-term-ate-orl . |
| Open Datasets | No | The paper uses simulated data from a simple MDP and a sepsis simulator (Oberst & Sontag, 2019), but does not provide specific access information (link, DOI, or a formal citation for a publicly available dataset) for the data used for training. The provided GitHub link is for the authors' code and data for the experiments, not a general public dataset. |
| Dataset Splits | No | The paper does not explicitly provide specific training/test/validation dataset splits needed to reproduce the experiment. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., CPU, GPU models) used to run its experiments. |
| Software Dependencies | No | The paper describes the model architecture and estimation methods but does not provide specific version numbers for software dependencies or libraries (e.g., PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | For the Q function, we use a feed-forward neural network parameterized separately for each of treatment and control. Each network consists of two hidden layers with 128 and 64 features respectively with sigmoid activation functions and a linear final layer with no activation function. Additionally, we maintain separate target networks by freezing the parameters of each network for 64 epochs... Table 1 in the Appendix lists the parameter values used in the experiments. |