2017-07-11 - Issue with Project Discovery Evaluation Set


(Moksa Elodie) #228

I hope that any obvious issue samples will get the accuracy score reallocated

Sample #200059672

(Moksa Elodie) #229

I just had Sample #200077053 and it was practically identical in the issue as your sample #200077052

(Hoshi) #230

200077055 and 200077050 are also the same star as those 2 above, considering the numbering 51 and 54 is probably also the same.

(Vincent Athena) #231

I had one of those “hard” ones, where the solution looks like noise. But I succeeded! My experience may be of use to others, and point to ways to improve the game.

I found one dip, marked it. Then found another, marked and and folded. It appeared I did get it right, but I noticed other dips off to the side, like I had the wrong period. So I discarded. That is when the pain began.

I had trouble finding my first dip, and finding the right period. This is because hitting “discard” starts you right at the beginning. I tried several times, different dips, folding discarding, until I got it right. Once I had it, all the dips folded on top of each other, it looked obvious. Until then, it was all hidden in the noise. Once I submitted and got “analysis success” I saw the complete data set with several groups of yellow dots that looked totally unremarkable, just regularly spaced sections of noise. It was only with folding that the results became easy to see.

Suggestion: we need a “back” button, or an “unfold” button, so we can keep our first mark, and try a vastly different period. Or, we need something like this

You can click and hold in the scroll box to slowly or quickly change the fold period. Click-and-hold is far easier than repeated dragging for large changes. Keep the dragging, though. Its good for final adjustments. Also, if you think you are off by a factor of two or three (as I was) you can test that as well by clicking the lower buttons.

Finally, we need to be able to fold after we get “Analysis failed” so we can see what success would have looked like.

(Tedric) #232

My tool wishlist:

  1. Magnifier glass (incoming).
  2. Ability to change the colour of the sample
  3. Basic averaging tools to toggle on and off (1,2,4,8,16h weighted).
  4. An Unfold button.
  5. When in folded mode, some kind of indicator to show where your “starting” fold length is (say a dot at the bottom of the folded window for each folded overlay).
  6. Some way to “lock” the vertical axis to stop it autoscaling (in combination with something to allow you to override/alter the default scale.
  7. in folded mode, a way to divide and multiply by 2 or 3 the fold length.

Analyst rank 12, 66%.

(Commander Aeris) #233

So THESE dips with more than average below the mean are NOT transits:

While THESE rises ABOVE the mean ARE??:

Before CCP’s “fix” I used to get regular samples. Now after the fix I only get these kind of nonsensical samples.

Complete and utter trolling tbh

Better to roll alts and just do the tutorial and level 1-10 a dozen times to get some skins…

(Vincent Athena) #234
  1. Basic averaging tools to toggle on and off (1,2,4,8,16h weighted).

This. Also, we should be able to display TWO averages at the same time, like a 1 hour average, and a two sample average. The two sample wide averaging window reduces noise without hiding the dip, and the longer term average gives a reverence line to see the dip against.

(Icarus Chelien) #235


Is there a forum for discussing things like the above? I’m not sure how the left one can be a transit and the second one isn’t?

(Shpenat) #236

I just got this image. Failed exactly same.

(Tvashnar Crendraven) #237

I’d rather see it adjust itself automatically, logarithmicly based on period, so that it always seemed the same. It seems plenty quick around 0.5d, but terribly slow at 10d.

(Shpenat) #238

I second this. Most of the additional tools mentioned in this post would help a lot.

I would also add Fast Fourier Transform (FFT) tool for repetition pattern checking. It could help finding those short period repetitions.

(Dior Ambraelle) #239

I got a few of these too:


Notice the sample ID!

Also, what am I supposed to do with this?

It would be really useful if we could report broken samples, though that function could get out of control really fast.

(America Night) #240

I’m desperate with examples like those ones

(America Night) #241

Something better with the changes, but there are still many confusing answers…

(Galaxxis) #242

Another missed and rather extremely obvious planet. I’m really not sure how they’re missing these.

(Vulvona Ride-in) #243

i dont know how to post!

shame on me!

(Vulvona Ride-in) #244

ok, as far as im concerned, i feel like CCP didnt gave us all the tool to analyse the exo-planet transit in those graph.

Basically, they teach us to look for:

1- noticeble pitch / dip in the luminosity graph
2- “V” or “U” shaped dip
3- recognisable pitch / dip pattern
4- same pithc deep and width
5- do not get confuse between a real “dip” and the natural cycle of the graph

But, look at this picture… how can we evaluate this optic signature as being a “planatary transit” according to the above mentionned rule?

We are missing some tools to detect those kind of transit.

My feeling is that those transits only apprear in those “already” analysed graphs that are used to establish your % of accuracy. They serve as barometer.

Now if you noticed, every few new analysis, after completing your task, your %accuracy will either goes up (succes) or down (failure). This metirc should prevent you from doing poor analysis and claim your reward freely. The rewards you claim (exp + ISK) are based on your % accuracy level.

In between those % accuracy varication you recieved after analysis, you have to analyse few additionnals samples that wont change your % accuracy regardless of it being a succes or a failure. Those analysis are the true one they want the community consensus about and were prolly not already analysed by the scientists.

Now, in the reference graphs (the one that influence your %accuracy according to failure or succes), i feel like there is some transit that are evaluated with some advanced methods like avarage reduction in the signal over a short period of time (floating averange) or something like that. Unfortunatly, the tutorial enforced by CCP doesnt give us the tools and knowledge to identify those transits.

Rework your tutorial CCP or get rid of those samples.


This text will be hidden

(TheSmokingHertog) #245

Funny that most stuff mentioned in SiSi feedback is coming back in here a lot.

(Elassus Herron) #246

I had my accuracy up over 80% yesterday, then today got a chain of hard samples, one after the other, until I was back down to 60%. There is no experience that made me want to give up more effectively than one of those failure snowballs, and it seems like a lot of the eval samples were found using methods that aren’t available to us, or (as others have said) are obviously wrong. CCP, this really needs to get fixed, because we need to feel like the system we’re using is fair (ie, the players who just spam NO TRANSIT shouldn’t be getting to the rewards faster than the people who are trying to submit good data), and worth the effort.

@CCP_Lebowski, are you going to be giving exoplanets another bugfix pass any time soon?

(America Night) #247

Now i am in 94% of accuracy and 110 of analyst rank and level 2, is slow but good, thank for changes made. :smiley: