These forums have been archived and are now read-only.

The new forums are live and can be found at https://forums.eveonline.com/

EVE General Discussion

 
  • Topic is locked indefinitely.
 

Project Discovery: Challenging & Interesting Samples

First post
Author
Previous Topic Next Topic
Yoshi Himomura
State War Academy
Caldari State
#221 - 2017-05-26 22:43:13 UTC
Just found this thread, and figured I should repost from https://forums.eveonline.com/default.aspx?g=posts&m=6953050#post6953050

Essentially, I'm pretty sure that something is off on either how these are matched/graded, or how I'm understanding the exercise (despite being a generally sharp individual that has stained cells IRL before)

Images and greater explanation is in the original post. What does everyone think / what should I do? Am I just not understanding at all?
Spc One
The Chodak
Void Alliance
#222 - 2017-07-13 05:34:43 UTC
How can this fail ?
Image:
http://www.netsky.org/eve/AFailedPD13072017a1.png

it should be 100% correct.
Shocked
S1dy
Uplifting Infernal Paradise
#223 - 2017-07-13 11:18:01 UTC
Spc One wrote:
How can this fail ?
Image:
http://www.netsky.org/eve/AFailedPD13072017a1.png

it should be 100% correct.
Shocked


That's a Pulsar. The troughs are the regular diminishing of the pulsing star, not by a transit. A transit would cause a more rapid decrease.
Bones Prefect
The Scope
Gallente Federation
#224 - 2017-07-13 13:02:28 UTC
https://imgur.com/a/njahu

Maybe take another look at this one? Im okay with missing the yellow one, but dayum if thats not a transit....
Bones Prefect
The Scope
Gallente Federation
#225 - 2017-07-13 13:23:57 UTC  |  Edited by: Bones Prefect
Vanessa Celtis
The Scope
Gallente Federation
#226 - 2017-07-13 19:56:39 UTC
Bones Prefect wrote:
Oh.. come on!

https://imgur.com/a/CVTi6

You sure?

https://imgur.com/a/4w11z

Another one...

https://imgur.com/a/ASsCN



Yea, I got those too, probably exactly the same ones and also failed. The first one is hilarious !
Jon Greyburst
Dark Space Exploratory OG-2717
#227 - 2017-07-13 19:59:21 UTC
For them to be making these sort of screw ups these had to be assigned by computer rather than hand, which means that they found the dips through actual analysis and calculation. You can look at some of these and tell a human being didn't do them, they are far too obviously screwed up. If this is the case then there is no reason we should be expected to be able to eyeball their harder graphs and see them.
Vanessa Celtis
The Scope
Gallente Federation
#228 - 2017-07-13 20:01:28 UTC  |  Edited by: Vanessa Celtis
Here is another good one !

http://i.imgur.com/vRjX5HG.png
Raven Dallacort
Center for Advanced Studies
Gallente Federation
#229 - 2017-07-14 06:37:18 UTC
Not sure how I failed this one. If I get the 1st, 3rd and 5th correct, how is it possible to get the 2nd and 4th wrong?

http://imgur.com/a/7tiZ8

ID: 200109146
Moros Chaput
Aliastra
Gallente Federation
#230 - 2017-07-14 18:14:33 UTC
Raven Dallacort wrote:
Not sure how I failed this one. If I get the 1st, 3rd and 5th correct, how is it possible to get the 2nd and 4th wrong?

http://imgur.com/a/7tiZ8

ID: 200109146

Might be because the star is orbited by two planets and not one, although how we're supposed to tell the two apart is beyond me. Here's a similar one I encountered.


Here are some other peculiar ones I got:


Actually got one right! This one was slightly more obvious.
Close miss. It's so hard to match up the transits in noisy data. Folding isn't very useful here either.
Looks as though there would be another planet with a wider orbit, but I guess not.

Here's one I thought was just plain odd. Look at the width of the band: the brightness fluctuates considerably, but at the same time there are no discernible dips and the graph is relatively smooth.

I have plenty more, but I think the pattern is pretty obvious here. I'm willing to wager the dev(s) accidentally implemented some tests that just can't be solved. I doubt the scientists who actually look at this data eyeball everything and the noisy data we see here probably isn't meant to be solved through simple observation.

Although the majority of the data reveals no transits (excluding the proficiency tests), I've come across ~3 tests that have had 90%+ detection rates by the players and where the transits were very noticeable, so that's neat. It can get pretty tedious spamming the "no transit" button over and over again.
Vanessa Celtis
The Scope
Gallente Federation
#231 - 2017-07-14 18:46:30 UTC  |  Edited by: Vanessa Celtis
My stats:

- Reached level 50 and it becomes almost as AFK and repetitive.

- Accuracy Rating: 62 - 72 %, it oscillates in between those two values. As soon as I reach 73%, crap samples come in and it brings me down to 62%, then back up to 72... round and round. Sometimes its good to let it drop on purpose so that you get interesting samples. The higher the percentage accuracy the more crap samples you get (those crap samples are either noise or non-identifiable with a human eye).

- At this % level, I spam the "No Transit" button 9 times out of 10. Each 10 samples a good one come in and I can accurately analyse it.

- 1/5 good and obvious sample are thrown at me by the algorithm as poorly analysed where in fact the algorithm is obviously wrong.


So basically that's it !... Exoplanet Discovery discovered.
Good luck to everybody and stock up replacement PC mice, you gonna need them.

I am going back top Null ratting, its more interesting and more profitable.
Blade Darth
Room for Improvement
Good Sax
#232 - 2017-07-14 23:52:21 UTC  |  Edited by: Blade Darth
Vanessa Celtis wrote:
As soon as I reach 73%, crap samples come in
You mean crap evaluation samples?
I got a slight suspicion some of them are thrown in for s..its and giggles, to see if lab rats can pick up on something in a 25-day sample it would normally take 2 years worth of data to find.

Vanessa Celtis wrote:
Null ratting, its more interesting
ee... nope.

Vanessa Celtis wrote:
more profitable.
Probably. Making 100k after looking at "wtf is that" for 20 minutes is not exactly good income.

Omen Navy Issue Tutorial
Raven Dallacort
Center for Advanced Studies
Gallente Federation
#233 - 2017-07-15 05:04:15 UTC
Heres another interesting fail. I don't think it's technically possible to achieve over 90% accuracy in this program, unlike the previous cell staining one.

http://imgur.com/a/2kara

ID: 200077052
Raven Dallacort
Center for Advanced Studies
Gallente Federation
#234 - 2017-07-15 05:20:39 UTC
Vanessa Celtis wrote:

- Accuracy Rating: 62 - 72 %, it oscillates in between those two values. As soon as I reach 73%, crap samples come in and it brings me down to 62%, then back up to 72... round and round. Sometimes its good to let it drop on purpose so that you get interesting samples. The higher the percentage accuracy the more crap samples you get (those crap samples are either noise or non-identifiable with a human eye).


I've noticed the exact same thing. I get close to 70% and then I get tossed samples that I have no clue where the transit is, so i'm continually wrong until I get closer to 60% and then it gives me samples I can "accurately" mark. Definitely something very wrong with the algorithm - it shouldn't be punishing you for being more accurate, but that's what it seems to feel like.
Yueh Ferrda
Doomheim
#235 - 2017-07-15 06:27:23 UTC  |  Edited by: Yueh Ferrda
It just a big SH**. They admit it during the live but do nothing for it. I've decided to falsify all data and go to 0.0% as CCP want.
Kevin Forrest
The Scope
Gallente Federation
#236 - 2017-07-15 08:07:46 UTC  |  Edited by: Kevin Forrest
I think this thing needs another wipe, but at the same time they should place a system in where you only get certain samples during every 5/10 levels, im level 63 on this atm and id love to get samples which havent been touched by a level 2 player who has no clue on what he's looking for and somehow inputs a pattern on a straight line

I've sent CCP a coouple of tweets asking for a reset, some of these samples are like trying to find a fart in a Jaccuzi but im sure there is no transits in half of them, im sure some of them i am wrong anyway but im losing far too much accuracy on some that even after zooming in, im still not seeing the pattern that i apparently missed. I believe a reset now after a few days of people getting the hang of it would actually provide better data in the long run.

WTF???
Blade Darth
Room for Improvement
Good Sax
#237 - 2017-07-16 01:07:03 UTC
Yueh Ferrda wrote:
It just a big SH**. They admit it during the live but do nothing for it. I've decided to falsify all data and go to 0.0% as CCP want.
geez.... someone ran out of Caldari Navy Fluoxetine?

Omen Navy Issue Tutorial
Annika Inari
Unicorn Rampage
#238 - 2017-07-16 06:53:19 UTC
The most "interesting" thing about this project is the ridiculous level of incompetence of the people making the evaluation samples. Half of them are flat-out wrong. I'm tired of being penalized because of their carelessness.
Emiko P'eng
#239 - 2017-07-16 11:27:07 UTC
Hehe!

Just starting out, so far the only problem I had up to now is spotting the small plants.

On those I just need to get my eye in on spotting the patterns Smile

Until I got this one?

It Does Not Like the Middle Dip?

I suppose it might be another planet but the fold looked perfect?
Larry Fat
Kruul TAX evasion
#240 - 2017-07-17 15:30:01 UTC  |  Edited by: Larry Fat
Okay I'll bite..

Came across this bullshit kek

All noise and a period you can't fit on screen.

ID:200060844