Real-Time Selection Under General Constraints via Predictive Inference

Authors: Yuyang Huo, Lin Lu, Haojie Ren, Changliang Zou

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate the breadth of applicability of the II-COS procedure by experiments on simulated data and real-data applications.
Researcher Affiliation Academia 1School of Statistics and Data Sciences, LPMC, KLMDASR and LEBPS, Nankai University, Tianjin, China 2School of Mathematical Sciences, Shanghai Jiao Tong University, Shanghai, China
Pseudocode Yes Algorithm 1 The data-driven II-COS procedure
Open Source Code Yes Code for implementing II-COS and reproducing the experiments and figures in our paper is available at https://github.com/lulin2023/II-COS.
Open Datasets Yes We consider the recruitment dataset from Kaggle [22] that contains 45,372 candidates... The other problem is to use 1994 Census Bureau dataset [8] to select a subset of individuals who may have high incomes in precision marketing.
Dataset Splits Yes we resort to a data-splitting strategy: randomly split historical data D into two parts, the training set Dtr and the calibration one Dcal of sizes n0 and n1 respectively. For each dataset, we randomly partition the data into three parts: ntr = 1,000 training data, ncal = 1,000 calibration data and the rest which are used as the online observations.
Hardware Specification Yes All the experiments were conducted on 3.11 GHz Intel Gen i5-11300H processors with 16 Gb memory at a Lenovo personal computer
Software Dependencies Yes R platform with version 4.2.1. implemented by R package nnet and R packages kernlab
Experiment Setup Yes As an example, we set the stopping rule as selecting total m = 100 samples, i.e., T = Tm = inft{t : Pt i=1 δi = m}. The predictor H is taken as random forest with defaulted parameters. We fix training data size ntr = 1, 000. Take α = 0.1 and K = 0.045 for FSR and m ES, respectively.