Julien Schueller
Julien Schueller
its mentioned there from the KL page: https://en.wikipedia.org/wiki/Proper_orthogonal_decomposition I dont think its related to incremental update
its meant to use the numerical value of range when computing the intersection with the distribution range and not the inf indication; I dont think we want to change that
it should be ot.HypothesisTest.ChiSquared, but yes its also wrongly exposed as ot.ChiSquared, will fix in misc 3
TODO: - remove the exception - parametrize by a minimum count of points per class (eg 10), then compute bin widths from quantiles
I misread the definition: Categorical was used when the support is the set of natural integers, but when the support is arbitrary in R they name it "DiscreteNonParametric": https://juliastats.org/Distributions.jl/stable/univariate/#Distributions.DiscreteNonParametric what...
if we can find a meaningful name
we should keep it, we can still use PenalizedLeastSquaresAlgorithm API directly with penalization if we wanted the issue is that the formula does not mention the penalization terms
of course, this must be fixed, done in misc2
input/output, per step and per failed/ok
this is the right thing to do unfortunately it cannot be exposed as the result is currenlty only exposed via the parent class, so adding methods wont be easy, thats...