Instances of the deprecated NumPy matrix class were replaced by instances of the NumPy array class.
1.0alpha
The article that the package comes with is finally out ! Update of the references in the functions' docstring.
0.9alpha
Add homoskedastic estimate for the ELMEnsembleRidgeCV class. As each ELM member of the ensemble is fitted with a different regularization parameter alpha obtained by Generalized Cross Validation (GCV), the effective degrees of freedom is computed considering each alpha separetely.
0.8alpha
Improvement of the computation of the degrees of freedom in the regularized homoskedastic case. The trace of P and P^2 is now computed via athe singular value of H, improving the computation time. A bug is also fixed in the Genelized Cross Validation computation of the ELMRidgeCV class.
0.7alpha
Add access to the gamma attribute used to compute the degrees of freedom in the regularized homoskedastic case
0.6alpha
Naive estimates added for comparison purposes, from "Model Variance for Extreme Learning Machine", F. Guignard, M. Laib and M. Kanevski, (in press), proceeding of the 28th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2020).