-
Notifications
You must be signed in to change notification settings - Fork 8
Known bugs
Karel Kubicek edited this page Mar 2, 2018
·
4 revisions
Following is the list of bugs we know but probably won't fix:
-
categories_evaluator
is producing biased p-values for random data. The likelihood of zero p-values should be 0, but it is around 5 % for common computation setup and this percentage is increasing with less iterations per epoch. This induces higher rejection rate of null hypothesis (data source is nonrandom) than is set by critical value alpha. Random-random experiment must be evaluated before interpretation of the results. Source of this bug is unconnected input and output in the circuit, which results in only one produced category in chi-square test. Therefore chi-square test produces zero p-value.
Bugs we want to fix are listed in project issues.