You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A common use case for independence is to use SurrogateTest and LocalPermutationTest to test for independence between two or more variables using some information theoretic (conditional) independence measure.
Currently, for both SurrogateTest and LocalPermutationTest, in the summary printout, it is checked whether the estimated measure m_est computed on the original data exceeds some percentile of the estimated measure computed over an ensemble of shuffled input data. Here's one example of such a print-out.
It would be nice to specify some threshold value that, if m_est < threshold, the interpretation printed to the user is always that we can't reject the null hypothesis. For example, with transfer entropy, we know that the value should be non-negative, so any values that are non-negative should immediately lead to rejection of the null.
It is up to the analyst to decide whether to do this, so threshold should be a keyword to the tests themselves. It can for example be a Real or a Nothing (in which case no threshold is applied).
The text was updated successfully, but these errors were encountered:
A common use case for
independence
is to useSurrogateTest
andLocalPermutationTest
to test for independence between two or more variables using some information theoretic (conditional) independence measure.Currently, for both
SurrogateTest
andLocalPermutationTest
, in the summary printout, it is checked whether the estimated measurem_est
computed on the original data exceeds some percentile of the estimated measure computed over an ensemble of shuffled input data. Here's one example of such a print-out.It would be nice to specify some
threshold
value that, ifm_est < threshold
, the interpretation printed to the user is always that we can't reject the null hypothesis. For example, with transfer entropy, we know that the value should be non-negative, so any values that are non-negative should immediately lead to rejection of the null.It is up to the analyst to decide whether to do this, so
threshold
should be a keyword to the tests themselves. It can for example be aReal
or aNothing
(in which case no threshold is applied).The text was updated successfully, but these errors were encountered: