I have updated the Lorentzian regularization that is applied in the DeepLDA CV to keep the output around 1.
Previously, in fact, it was not consistent with the manuscript (and with the same documentation in the example tutorial on DeepLDA), since the regularization was not applied to the final CV output, but to the NN output, before the LDA projection.
From a practical point of view, it does not change much as the two quantities differ by a linear combination; in fact, the notebooks show results virtually the same as before.
codecov[bot]
commented
8 months ago
I have updated the Lorentzian regularization that is applied in the DeepLDA CV to keep the output around 1.
Previously, in fact, it was not consistent with the manuscript (and with the same documentation in the example tutorial on DeepLDA), since the regularization was not applied to the final CV output, but to the NN output, before the LDA projection.
From a practical point of view, it does not change much as the two quantities differ by a linear combination; in fact, the notebooks show results virtually the same as before.