Learning Attribute Patterns in High-Dimensional Structured Latent Attribute Models
Authors: Yuqi Gu, Gongjun Xu
JMLR 2019 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The good performance of the proposed methodology is illustrated by simulation studies and two real datasets in educational assessments. |
| Researcher Affiliation | Academia | Yuqi Gu EMAIL Gongjun Xu EMAIL Department of Statistics University of Michigan Ann Arbor, MI 48109, USA |
| Pseudocode | Yes | Algorithm 1: PEM: Penalized EM for log-penalty with λ ( , 0), Algorithm 2: FP-VEM: Fractional Power Variational EM for Υ (0, 1], Algorithm 3: Stochastic Approximation Gibbs Screening |
| Open Source Code | No | The paper does not provide an explicit statement or link regarding the public availability of the source code for the methodology described. |
| Open Datasets | Yes | The good performance of the proposed methodology is illustrated by simulation studies and two real datasets in educational assessments...Example 1 presents a dataset from Trends in International Mathematics and Science Study (TIMSS)...Analysis of Fraction Subtraction Data. The fraction subtraction dataset is widely analyzed in the psychometrics literature (de la Torre and Douglas, 2004; De Carlo, 2011; Henson et al., 2009; de la Torre, 2011). |
| Dataset Splits | No | The paper analyzes real datasets (Fraction Subtraction Data, TIMSS) but does not specify any training, test, or validation splits. For simulation studies, it mentions generating data for specific sample sizes (e.g., N=500, 1000) and performing 200 independent runs, but does not describe splits within these generated datasets. |
| Hardware Specification | Yes | All the algorithms are implemented in Matlab. In particular, in the case of relatively strong signal with 1 θ+ j = θ j = 0.10, screening and computing an entire solution path for (2K, N) = (220, 1000) takes < 2 minutes on average on a laptop with a 2.8 GHz processor |
| Software Dependencies | No | All the algorithms are implemented in Matlab. No specific version of Matlab or other software dependencies with version numbers are provided. |
| Experiment Setup | Yes | In each scenario we perform 200 independent replications. For shrinkage estimation, we apply the proposed Algorithm 1 Penalized EM (PEM) and Algorithm 2 Fractional Power Variational EM (FP-VEM) , and also apply the plain EM algorithm with thresholding for comparison. When running PEM we compute a solution path by varying λ in the range of λ { 0.2, 0.4, . . . , 3.8, 4.0}, and select the λ that gives the smallest EBIC. When running FP-VEM we fix β = λ + 1 = 0.01 and compute a solution path by varying Υ in {1.0, 0.9, . . . , 0.4, 0.3} and also select Υ using EBIC. We use the threshold value ρN = 1/(2N) for the estimated proportions in the last step for all three shrinkage algorithms to select patterns (other smaller ρN values give similar results). |