Inference in a Partially Observed Queuing Model with Applications in Ecology

Authors: Kevin Winner, Garrett Bernstein, Dan Sheldon

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically validate the convergence behavior of our sampler and demonstrate the ability of our model to make much finer-grained inferences than the previous approach. We now report several experiments to evaluate the performance of our sampler and demonstrate the advantages of having a latent variable model. Our first two experiments empirically confirm the ergodicity result of Theorem 1 and demonstrate the improvement in mixing time resulting from adding supplemental moves to the sampler. Our third experiment demonstrates the running-time advantages gained by exploiting the log-concavity of the likelihood function within the sampler. We also provide a case study that compares the inference capabilities of our latent variable method compared to the previous approach of (Zonneveld, 1991).
Researcher Affiliation Academia 1College of Information and Computer Sciences, University of Massachusetts, Amherst, MA 01002, USA 2Department of Computer Science, Mount Holyoke College, South Hadley, MA 01075, USA
Pseudocode No The paper describes algorithms and methods in prose and mathematical equations but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any concrete access information for source code (no repository link, explicit code release statement, or mention of code in supplementary materials).
Open Datasets No The paper uses data generated from a probabilistic model with specified parameters rather than a publicly available dataset. For example, 'we generated data for a population of size N = 100 from a model with emergence density f S(s) Normal(µ = 8, σ = 4) and lifespan density is f Z(z) Exp(τ = 3) (parameterized by the mean τ) and computed observations at times t = {1, 2, 3, . . . , 20}'.
Dataset Splits No The paper does not provide specific dataset split information (e.g., exact percentages, sample counts, or citations to predefined splits) for training, validation, or test sets, as the data is generated synthetically.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library or solver names with version numbers, needed to replicate the experiment.
Experiment Setup Yes To evaluate the convergence of our Gibbs sampler under the hard constraints imposed when α = 1, we generated data for a population of size N = 100 from a model with emergence density f S(s) Normal(µ = 8, σ = 4) and lifespan density is f Z(z) Exp(τ = 3) (parameterized by the mean τ) and computed observations at times t = {1, 2, 3, . . . , 20}. We then performed MCMC from the intial configuration described in Section 4 using different subsets of the full move pool. Figure 3(a) shows the convergence of the cumulative mean negative log likelihood (NLL) of the first 1500 MCMC iterates. For this experiment, we fixed the parameters µ = 8.0, σ = 4.0, τ = 3.0, α = 0.5, t {1, 2, . . . , 20}, and varied the population size N to study the scalability of the sampler with respect to population size. Figure 4 shows the results, averaged over 10 trials for each value of N. The naive method scales approximately linearly (note that both axes are log-scale) with N, as expected, while the running time of the ARS-based algorithm grows very slowly with population size.