EVE General Discussion

For citizen science they're using an algorithm to determine the accuracy of humans...?

missing second planet in evaluation example:
http://i.imgur.com/Oe5zXAt.png

I have doubts about the theoretical basis for the analysis. We are given a month long intensity graph, and we are able to mark transits with a period fractions a of day. But how close to a star should a planet be and how fast should it rotate to have such period? For comparison, Mercury's orbital period is 88 days. I actually tried to estimate the distance from Sun (semi-major axis) for Mercury to have 1-day period. Since semi-major axis length is proportional to (orbital period)^2/3, 88^2/3 is roughly 20 - so Mercury would have to be ~20 times closer to the Sun, which seems too close. Not sure if my calculations are correct, though, not my field of study.

On the other hand, there's basically no way for us to distinguish between a small drop in intensity due to a transiting planet or due to a rotating sunspot(s) or some other process. However, I've found some info (like this article https://arxiv.org/pdf/1503.09042.pdf ) that several classes of stars have rotation periods of a few (1-7) days. So I tend to think that those control samples with lots of often questionable 'transits' are related to the star itself rather than an exoplanet. Which means that we probably shouldn't be able to mark anything with a period less than... um... 1-2 days or something? Or even more, if we are looking for planets, not just any periodic drops. Would save us some time when analyzing very noisy samples with no obvious dips.



As Blade Darth said, sudden jumps in intensity do this. It seems, the "detrend" option gives some kind of the derivative of the star intensity graph, and the derivative is sensitive to any sudden change compared to neighbor points. Without knowing the exact method they use to detrend, there is no way to tell for sure, but I'd be very cautios around rapid rises and falls since they seem to cause false positives in detrended graph.

They aren't hard. At this point, I can say with confidence that many of them are just wrong. It can't be in their best interest either. For a while now I thought, maybe they are right, and I'm overlooking something, so I started seeing transits in the tiniest irregularities, marking nonsense transits everywhere. But really, transits should be fairly rare, they said so themselves in the tutorial. And I suspect, if a sample looks like there is no transit, then there is probably no transit there, no matter what the evaluation sample says. Just ignore them, they are wrong.

And if they used an algorithm to create these evaluation samples - shame on them! I reckon they want proper data from us, so they could at least put in the effort to give us proper data in the first place.