Generalized Hierarchical Kernel Learning

Authors: Pratik Jawanpuria, Jagarlapudi Saketha Nath, Ganesh Ramakrishnan

JMLR 2015 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on benchmark REL data sets illustrate the efficacy of the proposed generalizations. Keywords: multiple kernel learning, mixed-norm regularization, multi-task learning, rule ensemble learning, active set method
Researcher Affiliation Academia Department of Computer Science and Engineering Indian Institute of Technology Bombay Mumbai 400076, INDIA
Pseudocode Yes Algorithm 1 Active Set Algorithm Outline Algorithm 2 Mirror Descent Algorithm for solving (14) Algorithm 3 Active Set Algorithm
Open Source Code Yes The implementations of both g HKLρ and g HKLMT ρ are available at http://www.cse.iitb.ac.in/~pratik.j/ghkl.
Open Datasets Yes We report the results of simulation in REL on several benchmark binary and multiclass classification data sets from the UCI repository (Blake and Lichman, 2013).
Dataset Splits Yes For every data set, we created 10 random train-test splits with 10% train data (except for MONK-3 data set, whose train-test split of 122 432 instances respectively was already given in the UCI repository).
Hardware Specification No The paper does not explicitly mention any specific hardware (e.g., CPU, GPU models, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions the implementation of algorithms but does not provide specific software dependencies or library versions used (e.g., Python, PyTorch, or other solvers with version numbers).
Experiment Setup Yes In each case, a three-fold cross validation procedure was employed to tune the C parameter with values in {10-3, 10-2, ..., 103}. For Rule Fit: All the parameters were set to the default values mentioned by the authors. In particular, the model was set in the mixed linear-rule mode, average tree size was set 4 and maximum number of trees were kept as 500. For SLI and ENDER: all parameters were set to their defaults.