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

This game is actually ridiculous. I was able to maintain perfect accuracy and do 5-6 samples per minute with the first iteration of PD. I did thousands of those slides because they were interesting and consistent and the baseline samples made sense. For this round, though, unless I spam “no transit” when it isn’t immediately obvious, I am down to two or three per minute and I cannot get above about 70% accuracy before tanking back down to 55 or 60 because of imaginary transits that are completely indistinguishable from random noise. This isn’t fun or worthwhile. I don’t really even care about the accuracy of my output anymore because if I can’t be provided with coherent and consistent samples I don’t think I’m going to return coherent and consistent data. There is literally no benefit from an accuracy standpoint for me to put effort into each slide when I can just continually spam no transit until I get an easy reference sample.

Do you want me to spam no transit? Because this is how you get me to spam no transit. This is one of the better ones.

2 Likes

While you would get a better line-up if you used the detrend tool, it probably still wouldn’t have made any sense.

I think you should include the full window in the screenshots as it shows the number of the sample in the lower right corner after failing.

Sample No: 200179159

These samples still appear all the time. How am I supposed to distinguish between the yellow marked area and anything else in that sample? (I used detrend, zoom etc.)

1 Like

Sample 200109146. I got that one as well. :confused:

1/1000th positioning near unlimited cycle times yeilds total online community annoyed. The nerf hammer is strong with this new mini game. So what devs will be dressing like Thor this Halloween?
Between this mini game or (hear me out) bashing my head on the desk with frustration… Fun factor is leaning towards the latter, so time to move keyboard.

Sample 200147923

I suppose in cases of slightly deviating orbital times you get such results. You can’t get them all in a perfect “line” because they aren’t in reality. Any possibility to add some modifier to let us tell you about it? @CCP_Leeloo

The interesting about this picture is the left side: there is something that looks like a transit, but isn’t marked.

Yes there are many cases like this. In the tutorial it says that most samples don’t include transits, so I go only with things that either look like a very specific “V” over a number of markers or just select “No Transits”. Reason being that most celestial objects should impact the light curve more than on just one marker. The only problem with this minigame is, that on many samples we have no way to distinguish between a “good” transit and just some other form of deviation.

Btw, this is another one I just got… that’s no fun :rage:

we need a probability algorithm or something that gives us a “Hey these markers might be a transit” on these super random level 5s.

Also the controls need to be re-evaluated.

It looks like the first transit you selected was towards the left side of the graph. The small error in the period accumulated through most of the graph. To limit this error, try to select the first transit as close as possible to the middle of the graph. This way, a small error in the period will accumulate through only half of the graph.

I will try that, thanks. In the other example I posted before, though, you can cleary see how the orbits are getting longer.

Just had another dafuq moment


So ok, I missed some dips, but the bigest dip at 2 days and 18 days aren’t?

Lets check the folded transits that are supposed to be there:


…dafuq?

2 Likes


Because the ‘peak’ at 24 days is a ‘bigger dip’ than at 26 days…

Please remove all the samples that have been analysed from samples longer than the 28 days we get to work with.

Yeah… not for human eyes. 1/3 of the dips are above the mean… as can be seen by the yellow peak in the middle of the dip, which makes it very hard to see a consistent dip in the sample.

As stated by others, this sample is 100% bugged. please reimburse my accuracy ~_~

  1. This is a really hard one
  2. The points are scattered all over the place, as in above and below the mean
  3. The interval is shorter than we can set for a single transfer interval… so either we need to set up two transfer intervals at 1/2 of each other or this is just too difficult with the current toolset

Edit > ok upon re-evaluation this isn’t shorter than the mininmum interval, however this does create a precedent for false-positives in the consensus samples.

Another really difficult one.

  1. My 2nd transit’s red dots overlap the yellow dots of a missed transit so they should’ve counted green/correct…
  2. No dip visible when checking the control fold.

If you look at your graph you’ll notice on the left side the transit markers are close to the right of the actual transit while the ones on the left are close to the left side. If you increased the orbit/interval time you would actually mark it. It’s not impossible, you just had the wrong orbit time set.

The interval is not shorter than half a day (although very close to 0.5) so you can actually mark it with one transfer interval. But that still doesn’t change the fact it’s pure luck to find transits like this.

1 Like