My target is to get samples from a complex distribution p(X,theta). Given: (1) p(X, theta) is not known, but we know p(X, theta) ∝ f(X,theta), where f is a known function; X is a random vector of N dimension; and theta is a parameter vector of K dimension. (2) theta is known. I want to get a ser…

You can use pm.Densitydist for both situations and just call pm.sample() - be aware that it might not be very efficient when you have large shape X.

Junpeng, Thanks for quick response. Great to know pymc3 can do it. But I still have problem on how to write the model. Let’s start from a one dimensional toy example: here I want to sample from a Bernoulli distribution with known p1=0.3. I typed the following codes: #sample from a bernoulli distr…

I read the post, but seems it does not answer my question. In the post, the purpose is to get posterior distribution of three parameters (m, q and sigmay). And four variables (‘logEiso’, ‘logEpeak_i’, ‘logErrEiso’, and ‘logErrEpeak_i’) are observed data. In the final model, priors are given to the …

It likely wont work out of the box with logp defined only in the discrete domain - likely you need to code up a special sampler. If your random variable is continuous on the real line (bounded continuous is fine but you will have to specify a transformation), here is an example: mu, sigma = 10., 2…

I copy your code to jupyter notebook and ran it. Next, I want to get the samples, so I ran: with m: step = pm.Metropolis() samples = pm.sample(1000, step=step) I got the error: AttributeError: Can't pickle local object 'loglike.<locals>.loglikefun'

You need to set ncore=1 as local functions are not pickleable: samples = pm.sample(1000, ncore=1)

Use: with m: samples = pm.sample(1000, ncore=1) Got: TypeError: function() got an unexpected keyword argument 'ncore' Use: with m: step = pm.Metropolis() samples = pm.sample(1000, step=step, ncore=1) Got: ValueError: Unused step method arguments: {'ncore'}

should be trace = pm.sample(1000, cores=1)

Thanks a lot. Two further questions: (1) Is it possible to do sampling on discrete variable, like Bernoulli variable in my original question? (2) In my research question, the random vector X contain both discrete and continuous components. In other word, it is in the form like X= [x1,x2], where x…

Can pymc3 get samples from a (not posterior) complex distribution using MCMC?

Questions

junpenglao June 30, 2019, 1:18pm 4

You are not using the Densitydist right, see eg: Problem model definition PyMC3

Topic		Replies	Views
How to set up a custom likelihood function for two variables Questions	8	13353	March 9, 2018
Can I avoid sampling irrelevant bits of the distribution? Questions	1	488	February 2, 2020
Create Custom Joint Distribution Questions	9	2641	May 26, 2019
How does PYMC sample a univariate distribution with no observed values? Questions doc	3	842	November 30, 2020
pm.DensityDist problem under PMYC 4 v5	17	2518	June 14, 2022

Can pymc3 get samples from a (not posterior) complex distribution using MCMC?

Related topics