MixtureDist Expression functionality

[Cygnus-A18]

Im having trouble structuring the logic for the expression functionality of the MixtureDist class, here is what I have currently:

class MixtureDist(Distribution):
    def __init__(self, *, name: str, coefficients: list, extended: bool, summands: list, pdf1: GaussianDist, pdf2: GaussianDist):
        super().__init__(name, "mixture_dist")
        self.coefficients = coefficients
        self.extended = extended
        self.summands = summands
    def expression(self, parameters):
        return coeff * pdf1 + (1.0 - coeff) * pdf2

Im looking at the following example:

{
      "coefficients": [
        "f"
      ],
      "extended": false,
      "name": "model",
      "summands": [
        "gx",
        "px"
      ],
      "type": "mixture_dist"
    }

Should I add two inputs which are both Dist objects to the mixturedist class init function?

[Cygnus-A18]

I've come to this but I am unsure of how to access the gx and px parameters separately

class MixtureDist(Distribution):
    def __init__(self, *, name: str, coefficients: list, extended: bool, summands: list):
        super().__init__(name, "mixture_dist")
        self.coefficients = coefficients
        self.extended = extended
        self.summands = summands
    def expression(self, parameters: dict(str, pt.scalar)):
        return parameters[self.coefficients[0]] * self.summands[0] + (1.0 - parameters[self.coefficients[0]]) * self.summands[1]

this is being called by:

for dist in distributions:
            print(dist)
            self.distributions[dist.name] = dist.expression(self.parameters)

So I am unsure of how to pull the two different distributions, right now im just multiplying them by the strings "px" and "gx" which obviously doesn't make sense but I am unsure of what to substitute here.

[kratsg]

Looking at this, it's gotta be more generic: https://root.cern.ch/doc/master/classRooAddPdf.html which is similar to https://www.tensorflow.org/probability/api_docs/python/tfp/distributions/Categorical .

Given that you have the list of parameters, we might also want to pull in a list of distributions too (I assume it can't necessarily be recursive? maybe it can).

So we want something like

distributions[self.summands[0]]
distributions[self.summands[1]]

in a similar way. It didn't really occur to me that we'd need the actual distributions or pdfs passed in here, but I guess that can happen in some cases of the pdf composition, and we need to take that into account somehow.