Polynomial time algorithms for dual volume sampling

Authors: Chengtao Li, Stefanie Jegelka, Suvrit Sra

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We report selection performance of DVS on real regression data (Comp Act, Comp Act(s), Abalone and Bank32NH1) for experimental design. We use 4,000 samples from each dataset for estimation. We compare against various baselines, including uniform sampling (Unif), leverage score sampling (Lev) [30], predictive length sampling (PL) [45], the sampling (Smpl)/greedy (Greedy) selection methods in [43] and Fedorov’s exchange algorithm [20]. We initialize the MCMC sampler with Kmeans++ [5] for DVS and run for 10,000 iterations, which empirically yields selections that are sufficiently good. We measure performances via (1) the prediction error ky X ˆ k, and 2) running times. Figure 1 shows the results for these three measures with sample sizes k varying from 60 to 200. Further experiments (including for the interpolation β < 1), may be found in the appendix.
Researcher Affiliation Academia Chengtao Li MIT ctli@mit.edu Stefanie Jegelka MIT stefje@csail.mit.edu MIT suvrit@mit.edu
Pseudocode Yes Algorithm 1 Markov Chain for Dual Volume Sampling
Open Source Code No The paper does not provide a statement about releasing source code or a direct link to a code repository for the methodology described.
Open Datasets Yes We report selection performance of DVS on real regression data (Comp Act, Comp Act(s), Abalone and Bank32NH1) for experimental design. We use 4,000 samples from each dataset for estimation. Note: The footnote links to http://www.dcc.fc.up.pt/?ltorgo/Regression/Data Sets.html.
Dataset Splits No The paper mentions '4,000 samples from each dataset for estimation' but does not specify explicit training, validation, or test splits by percentage, count, or reference to standard splits.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as CPU/GPU models or memory specifications.
Software Dependencies No The paper does not list specific software dependencies with version numbers needed for replication.
Experiment Setup Yes We initialize the MCMC sampler with Kmeans++ [5] for DVS and run for 10,000 iterations, which empirically yields selections that are sufficiently good. We measure performances via (1) the prediction error ky X ˆ k, and 2) running times. Figure 1 shows the results for these three measures with sample sizes k varying from 60 to 200.