Gaussian process with 10 datapoints

This arXival (accepted in Icarus) proposes to offer insights into the relative support for a periodic (seasonal) signal in Martian methane levels with respect to an aperiodic alternative.  The method adopted is to fit various flavours (kernels) of a Gaussian process regression and examine the posteriors.  With just 10 datapoints this is a very silly exercise since posterior coverage is nonsense and the sensitivity to prior choice incredibly high; neither of which are considered despite input from a professor who regularly uses and advocates GPs plus two referees.

(*) An aside: I once tried to convince the students of this professor that a wider class of stochastic process models could be considered beyond GPs (facilitated by sequential Monte Carlo methods for posterior sampling), but their response was that since a GP can fit anything it was all they needed for any job.  I needn’t go on.

This entry was posted in Uncategorized. Bookmark the permalink.

4 Responses to Gaussian process with 10 datapoints

  1. Tracy says:

    Who is the professor?

    • See acknowledgements in the arXival 🙂

      • Tracy says:

        Ah. These days exoplanet people seem very eager to throw GPs at everything. When you have a hammer, everything looks like a nail…?

      • I tend to throw GPs at a lot of things too, but lately I’ve been getting particularly nervous about what they’re actually doing; because of their bad frequentist properties and uncertainty about what they’re doing when the model is misspecified (which we can assume it pretty much always is in some way).

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s