The Inf-FS is a graph-based method which exploits the convergence properties of the power series of matrices to evaluate the importance of a feature with respect to all the other ones taken together. Indeed, in the Inf-FS formulation, each feature is mapped on an affinity graph, where nodes represent features and weighted edges relationships between them. Each path of a certain length l over the graph is seen as a possible selection of features. Therefore, varying these paths and letting them tend to an infinite number permits the investigation of the importance of each possible subset of features. The Inf-FS assigns a final score to each feature of the initial set; where the score is related to how much the given feature is a good candidate regarding the classification task. Therefore, ranking in descendant order the outcome of the Inf-FS allows us to perform the subset feature selection throughout a model selection stage to determine the number of features to be selected.
Reference : Infinite Feature Selection
Link Paper :http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=7410835
Why is the calculation of the A is different than the formula propoed in the paper?
This method is included in the FSLib2018. with demo file and examples. please see FSLib at https://uk.mathworks.com/matlabcentral/fileexchange/56937-feature-selection-library
how to run this code.
Hi, this is an interesting work! Just wondering if you have done similar work using PSO to select the features?
It is amazing~
+ Infinite Feature Selection Dec. 2016: "Unsupervised" & "Supervised" versions.
- Added new method: Features Selection via Eigenvector Centrality (ECFS) 2016
- New Inf-FS
- some problems fixed