Hi there, I am fairly new to pymc3 and was using it for some basic regressions to get started. So far I liked it a lot. My Problem is now getting a logistisc regression with weighted samples to run. For my problem one of the two classes is heavily undersampled and moreover some data points are more…

You are right, but from here on including the weights seemed straightforward to me. For completnes and if later users might have the same question here the final model with model: m = pm.Normal("m",0,sd=10,shape =len(cols)) x = df[cols].to_numpy() w = df["weight"].to_num…

Can you give us more details about what isn’t working, and/or some mock data which will reproduce the error? I got it to run with m = pm.Normal("m",0,sd=10) x = np.random.rand(200, 3) labels = np.random.randint(2, size=200) but I had dimension mismatches if I tried to include a shape in m.

Hi, thanks for the quick reply. In my example there are 8 inputs for each label, so the shape parameter is in fact 8. So i guess simple mock data for x could be: x = np.random.rand(200,8) But how did you come around using the shape parameter with more than just one x for each label? Omitting it d…

Apologies, I must have got mixed up between a couple of attempts before, and copied parts of two (don’t you hate when you’d swear you’ve done the same thing twice, and suddenly it’s not working anymore?). This worked for me: with pm.Model() as model: m = pm.Normal("m",0,sd=10, shape=8) x = …

I’m not sure on the weighting aspect but for labels try labels = df["label"].values

Thanks, this was it. I had it in in previous (not running) versions and I somewhere lost it. Now it runs as it should. Thanks a lot to both of you for your help.

Thanks for the solution guys! Just wanna point out that I’m not sure there is any weighting here, as w = np.ones(len(x)) is defined by never used in the model – am I missing something?

Perfect, thanks for including the full solution – you read my mind :wink:

Hey everyone, is there any follow up to this? Would you use pm.sample_posterior_predictive for this? I wonder how to eliminate the weighting for new examples.

How to run logistic regression with weighted samples

Questions

tomkov February 15, 2021, 2:25pm 10

how do you use posterior predictive distribution for new examples with this approach?

Pm.sample_posterior_predictive() not working with weights

Topic		Replies	Views
How to model/deal with weighted binary outcomes Questions	11	3127	August 30, 2021
How to perform weighted inference Questions	8	5333	August 30, 2021
Pm.sample_posterior_predictive() not working with weights Questions	12	1455	March 23, 2023
Sample posterior predictive error Questions	3	1231	October 8, 2019
Error in weighted inference model Questions	6	997	November 10, 2018

How to run logistic regression with weighted samples

Related topics