Open rendanim opened 4 years ago
Hi Rendani,
It's not unusual to get high NLPD and RMSE compared to the POE models that you're looking at. I mention in my paper that this happens even in the experimental results that I report because POE models are guaranteed to systematically underreport the predictive variance, although they can exhibit very good predictive point estimates. What values of K and J are you using here? Is it possible that you can plot the data in 1 dimension with the predictive mean and credible intervals to see if this is the case?
And yes, that is a local file. I will push it to the github repo.
Thanks,
Michael
Hi
Please could you indicate the expected results on the data set provided? Are high NLPDs and RMSEs expected on other regression data sets? Please see the ISMOE column below on UCI datasets. Is this expected?
Also, in online_ISMOE.Py the following import statement gives an error:
from mvn_t import log_mvn_t
Is this a local file?Thanks