CODEWITHSUNDEEP
pythonscipypymc3
Mike
Mike

Reputation: 341

Creating a shifted & scaled Beta in pymc3

I'm trying to create a beta RV that behaves like the one in scipy. Namely, it takes an alpha, beta, loc, and scale. I followed the discussion on creating a shifted gamma, and came up with something that looks like this:

import pymc3 as pm

class SSBeta(pm.Beta):
    def __init__(self, alpha, beta, loc, scale, *args, **kwargs):

        # Not sure what this is doing but it won't work without it. 
        transform = pm.distributions.transforms.lowerbound(loc)

        super().__init__(alpha=alpha, beta=beta, *args, **kwargs, transform=transform)

        self.scale = scale
        self.loc = loc
        self.mean += loc

    def random(self):
        return super().random()*self.scale + self.loc

    def logp(self, x):
        return super().logp((x - self.loc)/self.scale)

So I have two questions:

Upvotes: 3

Views: 435

Answers (1)

colcarroll
colcarroll

Reputation: 3682

An alternative is using pm.Deterministic. This does not allow you to pass observed data, but might be what you want anyways?

mu, scale = 2, 4
with pm.Model():
    b = pm.Beta('b', alpha=3, beta=8)
    shifted_b = pm.Deterministic('shifted_b', scale * b + mu)

enter image description here

Upvotes: 1

Related Questions