R2: An Efficient MCMC Sampler for Probabilistic Programs
Authors: Aditya Nori, Chung-Kil Hur, Sriram Rajamani, Selva Samuel
AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We also empirically demonstrate the effectiveness of R2 in particular, we show that R2 is able to produce results of similar quality as the CHURCH and STAN probabilistic programming tools with much shorter execution time. |
| Researcher Affiliation | Collaboration | Aditya V. Nori Microsoft Research Vigyan , #9, Lavelle Road Bangalore 560 001, India adityan@microsoft.com Chung-Kil Hur Department of Computer Science and Engineering Seoul National University Seoul 151-744, Republic of Korea gil.hur@cse.snu.ac.kr Sriram K. Rajamani and Selva Samuel Microsoft Research Vigyan , #9, Lavelle Road Bangalore 560 001, India {sriram, t-ssamue}@microsoft.com |
| Pseudocode | Yes | Figure 5: Given a statement S and a predicate ϕ defined over program variables, PRE(S, ϕ) is a pair ( ˆS, ˆϕ) where ˆS maps every sample statement with a pre-image predicate (via an observe statement immediately following the sample statement), and ˆϕ is a pre-image of ϕ over S. We assume that every while statement is annotated with a loop invariant ψ. |
| Open Source Code | No | The paper does not provide a link to its source code or explicitly state that the code for the described methodology is available. |
| Open Datasets | Yes | Letter1, Letter2: These are programs representing the probabilistic model for a student getting a good reference letter (Koller and Friedman 2009). Noisy OR: Given a directed acyclic graph, each node is a noisy-or of its parents. Find the posterior probability of a node, given observations (Kiselyov and Shan 2009). Burglar Alarm: This is an adaptation of Pearl s example (Pearl 1996) where we want to estimate the probability of a burglary having observed a phone call (see Figure 1). HIV: This is a multi-level linear model with interaction and varying slope and intercept (Hoffman and Gelman 2013). Chess: This is skill rating system for a chess tournament consisting of 77 players and 2926 games (Herbrich, Minka, and Graepel 2006). Halo: This is a skill rating system for a tournament consisting of 31 teams, with at most 4 players per team, and 465 games played between teams (Herbrich, Minka, and Graepel 2006). |
| Dataset Splits | No | The paper mentions using several benchmarks for evaluation but does not specify any training, validation, or test dataset splits or how data was partitioned for experiments. |
| Hardware Specification | Yes | All experiments were performed on a 2.5 GHz Intel system with 8 GB RAM running Microsoft Windows 8. |
| Software Dependencies | No | R2 is implemented in C++ and uses the Z3 theorem prover (de Moura and Bjorner 2008) in order to represent and manage pre-image predicates. |
| Experiment Setup | No | The paper does not provide specific details about the experimental setup, such as hyperparameters, optimization settings, or training schedules. |