EVE General Discussion

The lack of granularity is due to the telescope taking its measurements only so many times per day (seems to be around 150 times per day [estimate, not counting *all* those little dots]). Also due to the inconsistency in the light measurement due to God-knows what dust and space stuff (not to mention other photons) transiting the star in the light years between the telescope and the star.


I'm also getting very frustrated with this. I don't even care so much about the rewards, I mostly want to give them good data, and I can't do that if I can't learn how to give good data. =-( It would be nice if I could use the folding tool after I fail a slide.

EDIT: Yeah, I just cannot get the hang of this. I fail the known tests and I go back and loot at them and the transits I'm supposed highlight have upward spikes in them. That's the opposite of what's supposed to happen. Maybe I'll try again in a few months, let them iron out whatever problems are plaguing it. Or maybe I'm just not meant to be an exoplanet hunter. =(

What the!?

http://imgur.com/5BtZxpD

Edit:
K, I'm out for now. I'll check in a month or so again
http://imgur.com/ZbN82KI

Nah, that was a Successful Analysis.




Sometimes you will also find what are called False Positives that look like dips but are not.

First, I found a lot of people on here to be correct when they said it gets easier. Anytime I dip down to around 45% I get a few easy examples to push me back over 50% before getting some garbage again. A lot of times I can detect a faint pattern. I know something is there, but despite that knowledge I cannot pull out the answer they are looking for.

Second, the analysis UI/Tool is pretty but not practical. I can't exactly put my finger on what's needed but feel that the detrend option needs additional tuning parameters to allow us to better confirm/reject 'ghost' patterns. Don't disable folding after the answer has been revealed (let us explore and learn from errors)... Finally, I HIGHLY suspect that the data integrity of the training set may be flawed.... Maybe not, maybe we just need better tools, but I really do think some of the "answers" are indicating false positives. Once again, enable folding for exploration after the answer has been revealed.

Third, this is maybe the most frustrating, I'm pretty sure that a significant number of Eve players could do a better job if they were just given the raw data in a csv. Seriously, a moving average, some regression analysis, outlier detection and a simple check for autocorrelation could be done more effectively in Excel than through this UI... It's more frustrating than fun and doesn't advance the science. I'd be willing to bet that at least a few players more ambitious and enterprising than myself are already pumping the CREST data through R. Seriously, what are we doing here?


Edit - I've been playing a bit more and I am now convinced that the training set is flawed and that players are literally being trained to aggressively search for and flag false positives.

All you can say on the orbital period is that it is at least fourteen and a half days.

Also, just how. How?! How is someone supposed to tell there's a transit there? The yellow points look no different than the surrounding data points.

Agree. When I get a "failed analysis" I'd like to learn why.
For this to get successful, the reference sheets with known results should be extended with detailed data WHY the known result IS the known result. So we can learn.

Also, being able to fold/zoom after a failed analysis would be SUPER helpful.

I miss zooming on folds anyways.

Planets are Magick

I thought the game was trying to trick me, because the star is consistently getting brighter before and after the planet passes in front of it, and planets don't make stars brighter, short of falling into them.

So, if this really is a planet, how is the star getting brighter? The only thing I can imagine is gravitational lensing, and I'm not at all convinced that's a viable explanation.

Edit:

I decided to do the tutorial again, and it seems there really is a brightness increase. The things I don't pay attention to.

I have been grinding away at this for a few days now and seem to be doing OK. My accuracy is hovering around 70% and I have the hang of the tools available.

What confuses me are the results I get after I submit my findings.

Analysis succeeded/failed seems to mean that I did or did not match a professionally analyzed result. The impact is a bump or drop in my accuracy rating. This kind of makes sense but most/all of my "fails" are plates that I still can't see a pattern in and that is frustrating. There are also plates where I find some but not all or too many planets and that is counted as a fail instead of partial credit.

The consensus ones really make me wonder what we are doing. They don't hurt my accuracy score, but my results and what other people are reporting are very often wildly different. I tend to only mark what I consider an obvious single drop in brightness or in folded trends, but it appears to me that some folks are marking any dip below average which would create a bunch of false positives for the real researches to spend time reviewing. Or maybe they are right and I just can't see the patterns they found.

Anyway, I agree there is room for improvement, but I am going to keep doing it until I get really bored or burned out.

This is a joke, right? https://imgur.com/a/fWgnT

It doesn't. They put in some *supposedly* known ones to check if we are actually doing what we are supposed to be doing, and not just clicking around randomly. With the cells this worked for the most part, but with the exoplanets I have the feeling they have either not checked these test samples themselves, and let an algorithm do it that isn't very accurate, or we are getting the wrong samples for some reason.