Toolbox for Multimodal Learn (scikit-multimodallearn)

Authors: Dominique Benielli, Baptiste Bauvin, Sokol Koço, Riikka Huusari, Cécile Capponi, Hachem Kadri, François Laviolette

JMLR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Figure 1 shows the results of a benchmark, with 5-folds cross validation on the multi-view version of MNist presented in Section 2.2. We tested a mono-view decision tree and Adaboost (Schapire, 2013) on each view, their early fusion versions and the four algorithms of scikit-multimodallearn. All the implemented algorithms output higher accuracy scores than the mono-view approaches and the naive fusion methods. Mu Combo, despite being dedicated to imbalance problems, still displays an interesting score for this balanced task.
Researcher Affiliation Academia Dominique Benielli firstname.lastname[at]univ-amu.fr C ecile Capponi firstname.lastname[at]lis-lab.fr Sokol Ko co firstname.lastname[at]mines-stetienne.fr Hachem Kadri firstname.lastname[at]lis-lab.fr Riikka Huusari firstname.lastname[at]lis-lab.fr Department of Computer Science Aix-Marseille University, CNRS, LIS, 13013 Marseille, France Baptiste Bauvin firstname.lastname[at]lis-lab.fr Fran cois Laviolette firstname.lastname[at]ift.ulaval.ca Pavillon Adrien-Pouliot, Local PLT-3908, 1065, av. de la M edecine, Universit e Laval, Qu ebec (QC) G1V 0A6, Canada
Pseudocode No The paper describes algorithms (MVML, lp-MKL, Mumbo, Mu Com Bo) in text and provides code snippets for library usage in Python, but it does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes scikit-multimodallearn is a Python library for multimodal supervised learning, licensed under Free BSD, and compatible with the well-known scikit-learn toolbox (Pedregosa et al., 2011). If needed, the package can be downloaded from Github.2 The development has been performed using continuous integration with Docker and automated tests, covering 90% of the code. Footnote 2: Hosted here https://github.com/dbenielli/scikit-multimodallearn .
Open Datasets Yes In the following example, a multiview data set is instantiated with a multi-view version of MNist (Deng, 2012) for which we generated the HOG (Triggs and Dalal, 2005) in 12 directions, and selected 3 random directions for each view.
Dataset Splits Yes cross val score can be used transparently on multimodal estimator for multiviews data. >>> # usage cross_val_score on Mu Combo Classifier >>> cross_val_score(clfm , XX , y, cv=5) array ([0.96666667 , 0.96666667 , 0.9, 0.93333333 , 1. ]) Figure 1 shows the results of a benchmark, with 5-folds cross validation on the multi-view version of MNist presented in Section 2.2.
Hardware Specification No The paper does not provide any specific hardware details (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions software dependencies like "scikit-learn" and "numpy" but does not specify their version numbers. For example, Section 1 states: "Compatible with scikit-learn and numpy libraries".
Experiment Setup Yes >>> base_estimator = Decision Tree Classifier (max_depth =4) >>> clf = Mumbo Classifier ( base_estimator =base_estimator , n_estimators =4, random_state =7) >>> est = One Vs One Classifier (MVML(lmbda =0.1 , eta=1, nystrom_param =0.2)) >>> param = { estimator__learn_A : (1, 3), estimator__learn_w : (0, 1), estimator__n_loops : (6, 10), estimator__nystrom_param : (1.0 , 0.3) , estimator__kernel : ( linear , polynomial ) , estimator__lmbda : (0.1 ,), estimator__eta : (1,)}