I want to add some noise to the variable \alpha like this: \alpha=\alpha*exp(\varepsilon) where \varepsilon~Normal(0,\sigma) is a separate, independent noise term (non-trainable) and \sigma~HalfNormal(sigma=1) (trainable). The distribution of random variables in pymc is trainable as default, bu…

You need to implement a custom step sampler that always updates the value of the variable with a new draw from the prior, regardless of the model logp. This thread may be useful: Custom sampling step - #32 by ricardoV94

How to get a non-trainable random variable?

Questions v3

AlexF April 17, 2023, 1:39pm 4

Hi @wmylxmj!
I’m trying to implement something similar. Could you share some part of your code that solved the problem?

Topic		Replies	Views
Custom sampling step Questions	36	2637	November 8, 2022
Sample method of pymc3 Questions	4	685	July 15, 2018
Call MvNormal.random() method inside model Questions	3	1159	August 29, 2018
How to use free random variables in step method function? Questions	5	2281	July 30, 2018
Combining Theano variables and Numpy arrays in custom step function Questions	4	510	October 30, 2020

How to get a non-trainable random variable?

Related topics