Hello, I recently purchased Allen Downey’s Think Bayes - I’m working through Chapter 20 - Counting Cells (the chapter can be opened in Google Colab here ). There is a question I’m pondering that affects some of my models. It seems that we do not want all random variables to be adaptive (i.e., have…

Thank you Christian. That thread with the updated API yields the following code, which will fix the shaker volume variables so they do not update: import theano.tensor as tt rand_stream = tt.random.utils.RandomStream() with pm.Model() as model: yeast_conc = pm.Normal("yeast conc", …

Haven’t looked through this particular example. so take this for what it’s worth. If shaker volume is measured, then you can simply use the measured values. No need for a parameter at all. The fact that volume is/was a random variable is not relevant if you have realized values of that variable (…

Thanks for the response! Yes, the shaker volume is measured (at 9.0), but there is measurement error, so the shaker1_vol variable is included in the model to add noise (~0.5%). So another way to frame my question is whether there is a typical way to add a set amount of noise to a model. From my u…

[image] GBrunkhorst: …the shaker1_vol variable is included in the model to add noise… So I guess it depends on what exactly you mean by this. Are the measurements assumed to reflect/include this noise? If so, how do you know the noise is normally distributed? How do you know what the mean …

Sorry for the slow response - I think you would need to look at the Downey example for this to make sense (Colab link above). He is fitting a single data point with a bunch of priors based on the precision of laboratory procedures, and the goal is to calculate the error. I think that the pm.Normal…

Could be. But I read the model as implying that the volumes of the shakers and the samples from each shaker are uncertain quantities. So the “specification” (prior) suggests that these quantities are normally distributed with a specific mean and SD, but the modeler is open to the possibility of da…

I still have not figured out if it is possible to “freeze” a distribution so that the posterior is the same as the prior. However, I explored how a careful specification of priors can help. I added a notebook here.

Frozen or non-adaptive distribution for random variable

Questions

cluhmann January 3, 2022, 3:04am 9

Check out @junpenglao’s answers in this thread. Does that give you what you need?

Topic		Replies	Views
Prevent prior from updating? Questions	9	2126	February 12, 2023
Non-adaptable distribution from an other model posterior v5 modeling	6	115	October 16, 2025
Custom sampling step Questions	36	2763	November 8, 2022
How to perform ADVI only on some of the variables in a model? v5	6	407	August 8, 2023
Two-stage Bayesian regression enforcing a fixed distribution (not Just Hierarchical regression) v5 modeling , aesara	18	1668	August 5, 2022

Frozen or non-adaptive distribution for random variable

Related topics