Projection Pursuit Density Ratio Estimation
Authors: Meilin Wang, Wei Huang, Mingming Gong, Zheng Zhang
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that our proposed method outperforms existing alternatives in various applications. In this section, we compare our proposed projection pursuit density ratio estimation (pp DRE) method with existing alternatives using experimental and real-world data. |
| Researcher Affiliation | Academia | 1Center for Applied Statistics, Institute of Statistics and Big Data, Renmin University of China, Beijing, China 2School of Mathematics and Statistics, University of Melbourne, Melbourne, Australia. Correspondence to: Zheng Zhang <EMAIL>. |
| Pseudocode | Yes | The complete process of our approach given tuned hyper-parameters is summarized in Algorithm 1. Algorithm 1 pp DRE |
| Open Source Code | No | The paper mentions utilizing existing open-source implementations for baseline methods (e.g., 'Python densratio package for u LSIF', 'official open-source code3' for f DRE), and states that the authors 'implemented the algorithm' for RRND. However, it does not provide any explicit statement or link for the open-sourcing of their proposed 'pp DRE' methodology. |
| Open Datasets | Yes | We apply our proposed pp DRE method to estimate continuous treatment effects in the framework of Ai et al. (2021)... based on a semi-synthetic data set... IHDP-continuous dataset by leveraging the real-world covariates from the original IHDP dataset (Hill, 2011). In this section, we introduce the basic information and the regression tasks of the benchmark datasets that are used in the covariate shift adaption experiments...Abalone The Abalone dataset, accessible online4, is a popular dataset used in machine learning and statistics to predict the age of abalone from physical measurements. |
| Dataset Splits | Yes | For our proposed pp DRE method, as mentioned in Remark 4.3, the optimal hyperparameters are identified by the minimal validation loss in CV. For clarity, we detail our approach as follows: across all experiments, we utilized 5-fold CV with random sampling. |
| Hardware Specification | Yes | The experiments were carried out on a computing node utilizing dual AMD EPYC 7713 processors, providing a total of 128 CPU cores. |
| Software Dependencies | No | The paper mentions software like 'Python densratio package' and 'Light GBM model' for baseline methods, but does not provide specific version numbers for any of the software or libraries used in their experiments. |
| Experiment Setup | Yes | For our proposed pp DRE method, as mentioned in Remark 4.3, the optimal hyperparameters are identified by the minimal validation loss in CV. For clarity, we detail our approach as follows: across all experiments, we utilized 5-fold CV with random sampling. A grid search was conducted over a predefined set of parameter ranges, which are outlined in Table 4. Table 4. Search Grid for pp DRE Parameter Description Search Space: 𝐾 Number of PP iterations {5, 10, 15}, 𝐽𝑘 Number of basis functions {20, 50, 70, 100, 150}, 𝜆 ℓ2-regularization strength {0.5, 1, 5, 10}, 𝛿 Gradient descent learning rate {0.001, 0.01, 0.1}. |