Survival Permanental Processes for Survival Analysis with Time-Varying Covariates
Authors: Hideaki Kim
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our algorithm on synthetic and real-world data, and show that it achieves comparable predictive accuracy while being tens to hundreds of times faster than state-of-the-art methods. |
| Researcher Affiliation | Industry | Hideaki Kim NTT Human Informatics Laboratories NTT Corporation hideaki.kin@ntt.com |
| Pseudocode | No | The paper describes algorithms but does not provide structured pseudocode blocks or sections explicitly labeled 'Algorithm' or 'Pseudocode'. |
| Open Source Code | Yes | We provide python codes to reproduce the results in this paper1. 1Code and data are provided at https://github.com/Hid Kim/Surv PP. |
| Open Datasets | Yes | We examined the validity of Surv PP against the benchmark models on two real-world survival data sets, Mayo Clinic Primary Biliary Cholangitis data (PBC) and Standard And New Antiepileptic Drugs study data (SANAD), provided by R packages survival (LGPL-3) [41] and joine R (GPL3) [35], respectively. ... We also examined the validity of Surv PP against GBM on MIMIC-III Clinical Database (MIMIC III), the large publicly available dataset of over 50,000 ICU admissions from the Beth Israel Deaconess Medical Center [19]. |
| Dataset Splits | Yes | For each dataset, we randomly split the U individuals into 10 subgroups, assigned one to test and the others to training data, and conducted 10-fold cross evaluation of the predictive performances. ... In this experiment, we considered U = 103 and Ju {1, 5, 10, 20, 50}: the number of observed events N was 809 for λlin and 818 for λnon, respectively. ... We conducted additional experiments on independent validation datasets (see Appendix D.2), and on larger synthetic datasets (U 105) to examine the model s computation scalability regarding the event number N (see Appendix D.3). |
| Hardware Specification | Yes | A Mac Book Pro with 12core CPU (Apple M2 Max) was used, where GPU was set as off (tf.device( /cpu:0 )) for a fair benchmark comparisons. |
| Software Dependencies | Yes | For our proposal, we implemented Surv PP by using Tensor Flow-2.101. ... We implemented Cox PH and GBM with the established algorithms provided in the packages survival.coxph [41] and gbm3.gbmt [16], respectively; We implemented RFM by using class ltrcrrf in the open R code provided by Yao et al. [48]. |
| Experiment Setup | Yes | For GBM and Surv PP, the hyper-parameters were optimized through 9-point grid search: the number of trees and the shrinkage for GBM, and the kernel parameters for Surv PP. For details of the model configurations, see Appendix C. ... In Appendix C.1 (Synthetic Data - Surv PP): M = 500, lr = 0.05, G < 10 5. ... θ {0.1, 0.2, 0.5, 0.7, 1.0, 2.0, 5.0, 7.0, 10.0}. |