dfm / peerless

Single transit events in Kepler
MIT License
4 stars 4 forks source link

Is there a depth cut in your automated vetting? #14

Closed timothydmorton closed 8 years ago

timothydmorton commented 8 years ago

i.e., do you reject any transit/eclipse-like signals due to them having very large, obviously EB-like depths?

dfm commented 8 years ago

There isn't but I do have an impact parameter cut that will remove anything that looks V-shaped. Also, the completeness does seem to go down for very deep transits. I expect that this is mostly caused by the fact that I optimized my GP model for finding smaller transits but I'm not exactly sure.

timothydmorton commented 8 years ago

Ok I should add some more discussion about this (to explain why there aren't more EBs). Any idea how many signals you reject due to V-ness? On Sat, Jul 16, 2016 at 02:57 Dan Foreman-Mackey notifications@github.com wrote:

There isn't but I do have an impact parameter cut that will remove anything that looks V-shaped. Also, the completeness does seem to go down for very deep transits. I expect that this is mostly caused by the fact that I optimized my GP model for finding smaller transits but I'm not exactly sure.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/dfm/peerless/issues/14#issuecomment-233123021, or mute the thread https://github.com/notifications/unsubscribe-auth/ABzr29eELEXlxTOTcfBHFOuLLSMpW0zzks5qWKrzgaJpZM4JNyqi .

dfm commented 8 years ago

Yeah. Good idea.

I reject 2 candidates with the centroid offset test (these are both probably BEBs) and 7 candidates with the impact parameter cut (these ones are also probably astrophysical).

If I'm honest, one of the main reasons why I added the impact parameter cut was because it was a pain the in the ass to get the MCMC fits to converge on those candidates :-) but we could consider putting a table in with at least the transit times and depths for these candidates.

dfm commented 8 years ago

Also: two of these rejections are actually for the same system so I was double counting in that case. Only 6 systems were rejected by the vness cut.

dfm commented 8 years ago

Oops. One more mistake there. Those numbers were old, some of these were already found in the new EB catalog. After removing those, 1 candidate was rejected for the centroid motion and 3 were rejected for being V-shaped.

The real reason why there aren't very many EBs is because the recent catalog includes a table of long period EBs with only 1 or 2 eclipses and I removed all of those from the target list.

dfm commented 8 years ago

Reference: http://adsabs.harvard.edu/abs/2016AJ....151...68K

Look... Eric is an author :-)

ericagol commented 8 years ago

Yeah - I helped to find a bunch of heartbeat stars for that catalog.... Is this something you find as well?

Eric Agol Astronomy Professor University of Washington

On Jul 16, 2016, at 12:33 PM, Dan Foreman-Mackey notifications@github.com wrote:

Reference: http://adsabs.harvard.edu/abs/2016AJ....151...68K

Look... Eric is an author :-)

― You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

ericagol commented 8 years ago

I can't remember where I heard this idea, but someone suggested chopping each transit in half & fitting the first & second halves separately to see if you get the same parameters for both. This can help flag asymmetry for vetting.

Eric Agol Astronomy Professor University of Washington

On Jul 16, 2016, at 12:33 PM, Dan Foreman-Mackey notifications@github.com wrote:

Reference: http://adsabs.harvard.edu/abs/2016AJ....151...68K

Look... Eric is an author :-)

― You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.