![]() There is something I am not considering correctly, and I cannot see what it is. My doubt comes, when I want to make predictions, making use of the function “Predictive”, the results differ from the linear behavior that is being estimated. You can see in the example that the coefficients of the linear regression are much better estimated with the model that considers the errors. The example consists of a linear regression with a single independent variable, where errors in the measurement of “x” and “y” are considered.įollowing the suggestion you made above I consider the error of “y” as an independent node, which allows me to make the predictions and that these are not dependent on that observable. My doubt remains the same as in the first post, how to do the predictive analysis with the model and the “Predictive” function, or the alternative “Effect Handlers”. Hi are totally right, I was working with another example (one-dimensional) parallel to this one and I ended up mixing things up, sorry for the confusion. As it is emphasized in the text of the example and in the book, these coefficients do not differ much from each other, so I think that the prediction made with the same linear model (mu = a + bM * marriage + bA * age) in both cases cannot differ so much, as it is represented in the figure of the residuals.ĭivorce_sd = numpyro.sample(…, obs=divorce_sd)įrom my basic knowledge it looks good to me, so the model would be generative (and predictive) for “divorce_rate” and “divorce_sd”. The fact that this is a suspicious behavior can also be analyzed by comparing the regression coefficients obtained with “model 3” and “model 4”. This non-linear behavior (which in my opinion is not correct), explains the drastic decrease of the residuals with respect to “model 3”. Which is the evaluation used to make the predictions of “”. Predictions_4 = Predictive(model_se, samples_4)(rng_key_, marriage=, age=, divorce_sd=) The changes you suggest to make in the “model_se” do not change the behavior of “predictions_4”: ![]() I guess the result will be more linear if we replace I’m starting in the Bayesian world and I do it directly with numpyro, so if I get lost with something theoretical or technical even if it’s obvious, please forgive me. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |