Alternating optimization of decision trees, with application to learning sparse oblique trees

Authors: Miguel A. Carreira-Perpinan, Pooya Tavallali

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments show it consistently outperforms various other algorithms while being highly scalable to large datasets and trees. ... We constructed paths using initial CART trees of depths 4 to 12 (both axis-aligned and oblique) on the MNIST dataset, splitting its training set of 60k points into 48k training and 12k validation (to determine an optimal C or depth), and reporting generalization error on its 10k test points.
Researcher Affiliation Academia Miguel A. Carreira-Perpi n an Dept. EECS, University of California, Merced mcarreira-perpinan@ucmerced.edu Pooya Tavallali Dept. EECS, University of California, Merced ptavallali@ucmerced.edu
Pseudocode Yes TAO pseudocode appears in the supplementary material.
Open Source Code No The paper states, "The node optimization uses an ℓ1regularized linear SVM with slack hyperparameter C 0 ... implemented with LIBLINEAR [15]. The rest of our code is in Matlab." While it mentions using LIBLINEAR and Matlab, it does not explicitly state that the authors' own implementation code for TAO is open-source or publicly available.
Open Datasets Yes We constructed paths using initial CART trees of depths 4 to 12 (both axis-aligned and oblique) on the MNIST dataset... [27] MNIST. The MNIST database of handwritten digits. http://yann.lecun.com/exdb/mnist.
Dataset Splits Yes splitting its training set of 60k points into 48k training and 12k validation (to determine an optimal C or depth), and reporting generalization error on its 10k test points.
Hardware Specification Yes ran on a laptop with 2 core i5 CPUs and 12GB RAM
Software Dependencies No The node optimization uses an ℓ1regularized linear SVM with slack hyperparameter C 0 ... implemented with LIBLINEAR [15]. The rest of our code is in Matlab. The paper mentions LIBLINEAR and Matlab but does not provide specific version numbers for either.
Experiment Setup Yes The node optimization uses an ℓ1regularized linear SVM with slack hyperparameter C 0 (so the TAO sparsity hyperparameter in section 3 is λ = 1/C), implemented with LIBLINEAR [15]. ... We stop TAO when the training misclassification loss decreases but by less than 0.5%, or the number of iterations (passes over all nodes) reaches 14 (in practice TAO stops after around 7 iterations). ... We constructed paths using initial CART trees of depths 4 to 12 (both axis-aligned and oblique) on the MNIST dataset...