This function was developed by Oswaldo Ludwig. The arguments are the desired number of selected features (feat_numb), the matrix X, where each column is a feature vector example, and their respective target data y, which is a line vector. The output is a vector with the indexes of the features that composes the optimum set of features.