Hi everybody, I’m going to write a power-law distribution ( y = x^-2) in which x comes from a uniform distribution in [0,100] interval. According to this topic I could write the following code: import theano.tensor as tt def power_law(alpha): def logp(x): log_density = -alpha*tt.log(x) return…

That’s expected when not defining the interval transform so that the sampler can work on an unconstrained space. This seems to sample from the prior without divergences so it might work for you: import pymc3.distributions.transforms as tr def power(alpha): def logp(value): return tt.switch(…

Writing a new distribution (2)

Questions

fizikdan May 4, 2021, 10:03pm 9

Unfortunately, I couldn’t sample from your code!! It comes with big divergences (despite my target_accept which was 0.99!). Could you complete your code and sample from it?

Thank you

Topic		Replies	Views
Sampling from power law likelihood Questions modeling , sampling	4	762	April 18, 2023
Modelling: distribution conditioned on parameter being within interval Questions	1	422	February 18, 2019
Sampling uniformly in a triangular support Questions	15	2723	May 3, 2018
Bounded variable logp Questions	6	902	August 15, 2018
pm.find_MAP won't follow the bounds of custom distribution Questions	1	442	December 9, 2019

Writing a new distribution (2)

Related topics