Hyperbolic Random Forests
Authors: Lars Doorenbos, Pablo Márquez Neila, Raphael Sznitman, Pascal Mettes
TMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on standard and new benchmarks show that our approach outperforms both conventional random forest algorithms and recent hyperbolic classifiers. We extensively evaluate our approach on canonical benchmarks from previous works on hyperbolic classification. Additionally, we introduce three new multi-class classification experiments. We show that our method is superior to both competing hyperbolic classifiers as well as their Euclidean counterparts in these settings. |
| Researcher Affiliation | Academia | Lars Doorenbos EMAIL University of Bern Pablo Márquez-Neila EMAIL University of Bern Raphael Sznitman EMAIL University of Bern Pascal Mettes EMAIL University of Amsterdam |
| Pseudocode | No | The paper describes algorithms and methods using mathematical formulas and textual descriptions, but does not include any explicitly labeled "Pseudocode" or "Algorithm" blocks, nor does it present structured code-like steps. |
| Open Source Code | No | The paper mentions its review on Open Review and uses third-party tools like Horo SVM, but it does not provide an explicit statement or a link to its own implementation code for the described methodology. |
| Open Datasets | Yes | We evaluate Horo RF on two canonical hyperbolic classification benchmarks: Word Net subtree and network node classification (Cho et al., 2019; Fan et al., 2023). We run binary classification experiments on the nouns of Word Net (Fellbaum, 2010), embedded into hyperbolic space using hyperbolic entailment cones (Ganea et al., 2018a). We run node classification experiments on four network datasets embedded into hyperbolic space. These networks are (1) karate (Zachary, 1977), (2) polblogs (Adamic & Glance, 2005), (3) football (Girvan & Newman, 2002), and (4) polbooks1, a network of 105 books split into three affiliations. 1http://www-personal.umich.edu/ mejn/netdata/. We evaluate the methods using CIFAR10 Krizhevsky et al. (2009) and STL10 Coates et al. (2011). |
| Dataset Splits | Yes | The datasets are split into train, validation, and test sets with a ratio of 60:20:20. Following (Fan et al., 2023), we split the data into 80% for training and 20% for testing and report the best results from a grid search over three runs in AUPR. Following previous work (Fan et al., 2023), we run a grid search with 5-fold stratified cross-validation and report the best average micro-f1 score over five seeds, using a different embedding for each seed. |
| Hardware Specification | No | The paper discusses the runtime and computational time needed for Horo RF but does not provide any specific hardware details such as GPU models, CPU models, or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper refers to using specific methods like Horo SVM and mentions general concepts like Gini impurity, but it does not provide version numbers for any specific software libraries, programming languages, or solvers used in the implementation. |
| Experiment Setup | Yes | We use 100 trees for all tree-based methods. We use the Gini impurity to calculate the information gain. At every node, we set C to 2n where n is randomly sampled from { 3, 2, . . . , 5}. We stop splitting when a node reaches a certain number of samples m. In the case of ties in information gain between possible splits, a random split is selected. If the Horo Splitter fails to converge, we choose the best horosphere from a small number of random ideal points. We set m and the class-balancing hyperparameter β via grid search. A hyperparameter search is done on the validation set, and the macro-f1 score over three trials on the test set is reported. |