Deep-Treat: Learning Optimal Personalized Treatments From Observational Data Using Neural Networks
Authors: Onur Atan, James Jordon, Mihaela van der Schaar
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare our algorithm against state-of-art algorithms on two semi-synthetic datasets and demonstrate that our algorithm achieves a significant improvement in performance.Our results show that our algorithm achieves significant improvements with respect to strong baselines.Experiments on IHDP dataset for ITE estimation.Experiments on Breast Cancer Dataset for Policy Optimization. |
| Researcher Affiliation | Academia | Onur Atan,1 James Jordon,2 Mihaela van der Schaar2,3,1 1Electrical Engineering Department, University of California Los Angeles, USA 2Department of Engineering Science, University of Oxford, UK 3Alan Turing Institute, London, UK |
| Pseudocode | Yes | Algorithm 1 Deep-Treat |
| Open Source Code | No | The paper does not provide an explicit statement about releasing the source code for the described methodology, nor does it include any repository links. |
| Open Datasets | Yes | The semi-simulated dataset in (Hill 2011; Johansson, Shalit, and Sontag 2016) is based on covariates from a real randomized experiment...We evaluate our algorithm on a dataset of 10,000 records of breast cancer patients participating in the National Surgical Adjuvant Breast and Bowel Project (NSABP) by (Yoon, Davtyan, and van der Schaar 2016). |
| Dataset Splits | Yes | Within each iteration we randomly divide the IHDP dataset into training set (70%) and testing set (30%).We divide the other 80% of the data into 3 parts to train, validate and test the performance of the algorithms. |
| Hardware Specification | No | The paper mentions training models and using optimizers but does not provide specific details about the hardware used, such as GPU/CPU models or memory specifications. |
| Software Dependencies | No | The paper mentions 'Adam optimizer' and 'rectified linear unit' as an activation function but does not provide specific version numbers for any programming languages, libraries, or frameworks used (e.g., Python, TensorFlow, PyTorch). |
| Experiment Setup | Yes | We implement our algorithm with Le = Ld = 2 layers with 50 hidden units and 2 layers for potential outcome estimation for both treated and control groups with 50 hidden units with dropout 0.3.We evaluate our algorithm with Le = 1, Ld = 1 auto encoder layers (with 25 hidden units in the representation block) and Lp = 2 policy layers with 50 hidden units. We use relu as the activation function. We validate the hyper parameters of our algorithm on λ 1 10 3, 10 2, . . . , 102 .We train each algorithm with the Adam optimizer and with early stopping criteria. |