Closed ferrine closed 10 months ago
The parameterization of a default phi concentration is getting worse with number of dimensions, leading to multimodal default prior. Replacing the default with just ones, would not make the prior multimodal.
https://github.com/pymc-devs/pymc-experimental/blob/3444ede907abe415251f8ba28c70b670aa94c9f9/pymc_experimental/distributions/multivariate/r2d2m2cp.py#L196
Let's change the default? It should not be too late since it is experimental and anyway subject to change.
The parameterization of a default phi concentration is getting worse with number of dimensions, leading to multimodal default prior. Replacing the default with just ones, would not make the prior multimodal.
https://github.com/pymc-devs/pymc-experimental/blob/3444ede907abe415251f8ba28c70b670aa94c9f9/pymc_experimental/distributions/multivariate/r2d2m2cp.py#L196
Let's change the default? It should not be too late since it is experimental and anyway subject to change.