Hi, I am trying to adapt the example from the “Model Based Machine Learning Book”, chapter 3 ( link1 link2 ), where outcomes are parametrized as resulting from a deterministic greater-than comparison: [image] The model works okayish for the observed outcome=1 (there are many divergences), but it fa…

Yes you can also use a cumulative Gaussian in PyMC3: with pm.Model() as m: JSkill = pm.Normal('JSkill', 120., 40.) FSkill = pm.Normal('FSkill', 100., 5.) lg_p = pm.distributions.dist_math.normal_lccdf(JSkill - FSkill, pm.math.sqrt(50.), 0.) JWin = pm.Bernoulli('JWin', pm.math.exp(lg…

How to model a deterministic "greater than"?

Questions

junpenglao June 24, 2019, 9:40am 6

… Bad news to this is that if you see a lot of divergence the model is NOT working…

I really think you should change your model parameterization. See eg: Trueskill model in pymc3 - #2 by junpenglao

Topic		Replies	Views
How to model observed percentages (bounded from 0 to 1) Questions	8	2495	January 3, 2018
Question about how to force y_pred to positive numbers and advice seeking on my exercises v3 modeling	2	415	May 15, 2023
Bounding Output and Potential Values Questions	4	1438	June 20, 2018
How to condition on a parameter Questions	2	904	March 3, 2018
Log probability negative infinity Questions	4	1223	December 31, 2019

How to model a deterministic "greater than"?

Related Topics