Regularization in directable environments with application to Tetris

Authors: Jan Malte Lichtenberg, Özgür Şimşek

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Across a wide range of learning problems, including Tetris, STEW outperformed existing linear models, including ridge regression, the Lasso, and the non-negative Lasso, when feature directions were known. Our empirical analysis shows that these properties translate from the equal-weights model to STEW.
Researcher Affiliation Academia 1Department of Computer Science, University of Bath, Bath, United Kingdom.
Pseudocode Yes The pseudo-code is provided in the Supplementary Material.
Open Source Code No The paper mentions pseudocode in supplementary material, but does not provide an explicit statement or link for open-source code for the described methodology.
Open Datasets Yes We first consider the Rent data set (Tutz, 2011) where the problem is to estimate the response rent per m2 for 2053 apartments based on 10 features. In the Diabetes data set, in which a quantitative measure of disease progression of 442 diabetes patients needs to be predicted based on age, sex, body mass index, average blood pressure, and six blood serum measurements
Dataset Splits No The paper mentions tuning regularization strength using cross-validation, but does not specify explicit training/test/validation dataset splits (e.g., percentages or sample counts) in the main text.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory, or detailed computer specifications) used for running experiments were mentioned.
Software Dependencies No The paper does not specify software names with version numbers for libraries or tools used in the experiments (e.g., 'scikit-learn', 'PyTorch').
Experiment Setup Yes We used a board size of 10 10, with rollout parameters M = 7, T = 10. Multinomial logistic regression in iteration k used the most recent n(k) training samples, where n(k) = min(50, k/2 + 2). The regularization strength λ was tuned using cross-validation. Eight features were used to describe a state-action pair: landing height, number of eroded piece cells, row transitions, column transitions, number of holes, number of board wells, hole depth, and number of rows with holes.