How does Orange calculate confidence intervals in its Distribution widget? -
when using distribution widget of orange on binary classification dataset there's option of showing confidence intervals probabilities of given class label feature values, see: distribution widget doc
how these intervals calculated? i've tried searching the github repo using keywords: 'distribution', 'confidence interval'. have found code widget ui , no pointers actual stats calculated.
it's done in calchistogramandprobgraph method of owdistributions.py (code), code distributions widget.
for discrete features it's observed ratio. continuous features calls out c++ code (i assume) discretizes feature , estimates probability in similar fashion.
Comments
Post a Comment