Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

step_detect: remove ad-hoc restriction interval size>=2 #745

Merged
merged 3 commits into from
Sep 22, 2018

Conversation

pv
Copy link
Collaborator

@pv pv commented Sep 21, 2018

Allow length-1 intervals in step detection.

The Bayes criterion is supposed to select the interval cost correctly,
so this sort of ad-hoc restrictions are not necessary.

Also fix an unrelated performance issue with zero-variance inputs.

pv added 2 commits September 22, 2018 01:45
When variance is zero, use scale 1.0 for gamma. The actual solution is
always single interval, but this is the minimal correction required
in order to not look for solutions in silly places.
@pv pv force-pushed the step-detect-1 branch 6 times, most recently from 24a4281 to ef2e16d Compare September 22, 2018 17:38
Penalise 'just-so' overfitted sigma=0 models in the bayesian information
criterion, by fixing a noise floor in the prior.

Previously, these cases were usually not encountered because of
min_size=2, but with min_size=1 all datasets have an overfitted model
which should not be always accepted.
@pv pv merged commit 6f729bb into airspeed-velocity:master Sep 22, 2018
@pv pv deleted the step-detect-1 branch October 21, 2018 00:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant