Use LombScargleFast from gatspy? about cesium

[stefanv]

How hard would it be to update gatspy to calculate the same statistical quantities? Or, can we build those on top of gatspy? How does the output of gatspy compare to ours?

[bnaul]

I think it's mostly doable. The major things not included in gatspy are:

• Support for multiple frequencies --> easy to build on top
• Automatic selection of regularization parameter --> implementing this exactly the same way that ours works is probably not possible; however, achieving something functionally equivalent might be
• Statistical significance of fit --> easy to build on top
• Estimate linear trend --> would need to add to gatspy but this seems straightforward and like a sensible addition

My preference would be to ditch our implementation altogether since I think it is literally unmaintainable (I have spent hours looking at the code and about 50% of it is still totally inscrutable) and adds an additional layer of complexity (compiling Cython code) to our stack with little benefit.

[acrellin]

I strongly vote for replacing the TCP implementation with gatspy if possible. If we don't understand it and can't maintain it, we shouldn't be using it.

[stefanv]

I agree. If we can contribute missing functionality back to gatspy and
build the rest on top, that'd be excellent.
On Dec 4, 2015 12:12 PM, "acrellin" [email protected] wrote:

I strongly vote for replacing the TCP implementation with gatspy if
possible. If we don't understand it and can't maintain it, we shouldn't be
using it.

—
Reply to this email directly or view it on GitHub
#111 (comment).