Look Ma, No Latent Variables: Accurate Cutset Networks via Compilation
Authors: Tahrima Rahman, Shasha Jin, Vibhav Gogate
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluated the performance of cutset networks along two dimensions: (1) Model fit measured using the test set log-likelihood score and (2) MAP estimation quality measured using the log-likelihood of the MAP assignment. We used 20 benchmark datasets used in numerous prior studies... |
| Researcher Affiliation | Academia | 1Department of Computer Science, The University of Texas at Dallas, United States. Correspondence to: Tahrima Rahman <tahrima.rahman@utdallas.edu>. |
| Pseudocode | Yes | Algorithm 1 LC-CN (D,Q) Input : Training examples D defined over a set of variables X and a tractable latent model representing a distribution Q Output :A Cutset network |
| Open Source Code | No | The paper does not provide a direct link or explicit statement about the availability of the source code for the methodology described in this paper. |
| Open Datasets | Yes | We used 20 benchmark datasets used in numerous prior studies (Rooshenas & Lowd, 2014; 2013; Gens & Domingos, 2013; Vergari et al., 2015) to evaluate our new algorithm. All the datasets are defined over binary valued variables with the number of variables ranging from 16 to 1556 (see Table 1). |
| Dataset Splits | Yes | Table 1. Average test set log-likelihood scores. Bold values indicate best scores obtained by CN, CNx D, CNR, ACBN, ACMN or PSDD. Dataset Dataset Characteristics Test-set Log-Likelihood No Latent Variables Latent Variables #Var Train Valid Test... nltcs 16 16181 2157 3236 |
| Hardware Specification | No | The paper mentions 'in single CPU settings' and 'If parallel architectures or GPUs are used' but does not provide specific details on the hardware models (e.g., CPU, GPU, memory) used for the experiments. |
| Software Dependencies | No | The paper mentions the 'libra toolkit' and algorithms like expectation-maximization and Chow-Liu, but it does not specify software versions for reproducibility. |
| Experiment Setup | Yes | For all algorithms evaluated, we used a time bound of 48 hours and space bound of 4 GB. We used iterative deepening search for cutset networks, namely we start with a depth bound of 1 and progressively increase it until the time or space bound is reached. At each recursive call in Algorithm 1, we set the hyperparameter α (see Eq. (4)) using the validation set. We varied α from 0.0 to 1.0 in increments of 0.1. ... We varied the number of mixture components from 2 to 50 and bags from 2 to 40. We used a depth-bound of 5 in each bag for cutset networks. |