Software development for "Bayesian nonparametric population inference". In other words, just the direct application of probability theory to get the most general, principled, model-free inference we can have.
Creative Commons Zero v1.0 Universal
4
stars
1
forks
source link
Extract metainformation about boundary probabilities and quantiles of ordinal quantities #38
At the moment, the algorithm gives a prior probability of 1/8=12.5% to each of the boundary values of a bounded-domain variate (lines 176–177 of util_vtransform.R). This is arbitrary – but some choice had to be made.
In the case where the metadata are extracted from a (too) much larger dataset than the one that will be used in the algorithm, it makes sense to extract the prior probabilities mentioned above from this dataset instead.
Implement this.
Similarly for the quantiles of ordinal variates; this is partially implemented with useOquantiles.
This extraction of metainfo doesn't make sense, though, if it happens from the same data that will be used for the population inference. So maybe we should add a question to the user in buildmetadata about the two data sizes? Connects to #37 .
At the moment, the algorithm gives a prior probability of 1/8=12.5% to each of the boundary values of a bounded-domain variate (lines 176–177 of
util_vtransform.R
). This is arbitrary – but some choice had to be made.In the case where the metadata are extracted from a (too) much larger dataset than the one that will be used in the algorithm, it makes sense to extract the prior probabilities mentioned above from this dataset instead.
Implement this.
Similarly for the quantiles of ordinal variates; this is partially implemented with
useOquantiles
.This extraction of metainfo doesn't make sense, though, if it happens from the same data that will be used for the population inference. So maybe we should add a question to the user in
buildmetadata
about the two data sizes? Connects to #37 .