Figure 1
From: A framework for generalized subspace pattern mining in high-dimensional datasets
Overview of the GABi process. A diagram illustrating the GABi process. Starting with an input data matrix A, a population of candidate solutions S is initialized to S0. S, which encodes the bicluster columns I k for each candidate solution k, is iteratively updated through the GA loop. At each step of the GA loop, each solution is evaluated in turn: rule-based feature selection is applied to identify J k given I k and A (and potentially external information), then the fitness score for the solution (F k ) is calculated based on the function f(I k ,J k ). Based on the fitness scores F, candidate solutions are selected so that fitter solutions are more represented in the next iteration. Solutions are combined through crossover operations, and finally randomly ‘mutated’ before re-entering the solution population S as it goes through the GA loop again. When convergence criteria are met, each candidate solution’s I k are extracted and the corresponding J k identified through rule-based feature selection. The (I,J) sets encode the biclusters that are the output of the algorithm.