I was thinking that a parameterization number could be added as a (discrete) shape parameter.
The make_distribution documentation describes how a process_parameters method can be used to preprocess parameter when there are multiple parameterizations (e.g. see the MyBeta example). That is a little different because there we actually want the user to use different keywords to provide the parameters for the different parameterizations.
Here, I think we'd still use a and b (or alpha and beta, and maybe we'll have to add mu and delta) and a new parameterization (p?) shape parameter. There would just be one parameterization as far as the infrastructure is concerned. I don't remember the details of the Stable distribution parameterizations, but the process_parameters method might still be useful if the shape parameters need to be pre-processed (e.g. converted to equivalents in S0 parameterization).
There may need to be additional pre-processing of the argument x to methods like pdf and post-processing of the output of methods like icdf/sample/moment. I'm just suggesting that hopefully the moment
I was thinking that a parameterization number could be added as a (discrete) shape parameter.
The
make_distributiondocumentation describes how aprocess_parametersmethod can be used to preprocess parameter when there are multiple parameterizations (e.g. see theMyBetaexample). That is a little different because there we actually want the user to use different keywords to provide the parameters for the different parameterizations.Here, I think we'd still use
aandb(oralphaandbeta, and maybe we'll have to addmuanddelta) and a newparameterization(p?) shape parameter. There would just be one parameterization as far as the infrastructure is concerned. I don't remember the details of the Stable distribution parameterizations, but theprocess_parametersmethod might still be useful if the shape parameters need to be pre-processed (e.g. converted to equivalents in S0 parameterization).There may need to be additional pre-processing of the argument
xto methods likepdfand post-processing of the output of methods likeicdf/sample/moment. I'm just suggesting that hopefully the moment