Skip to content

Adaptive update interval #11

Open
Open
@kbarbary

Description

It can be expensive to determine bounding ellipsoids, particuarly for the multi-ellipsoidal method. However, this doesn't need to be done every iteration. The update_interval parameter lets the user set how often the ellipsoidal decomposition is performed.

In theory, fewer updates mean less efficient sampling and more likelihood evaluations. Thus, the ideal choice for this parameter depends on the runtime of the likelihood function versus the runtime of the ellipsoidal decomposition. It should be possible to time the execution of the likelihood function while sampling in order to adaptively adjust the update interval.

However, this might depend on having a good understanding of the effect on the sampling efficiency, as measuring the sampling efficiency directly might give results too noisy to be useful.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions