|
Author |
Thread Statistics | Show CCP posts - 4 post(s) |
|
CCP Wonderboy
C C P C C P Alliance
10
|
Posted - 2016.03.21 11:18:58 -
[1] - Quote
Hello scientists!
The Project Discovery team is extremely satisfied and impressed with Capsuleer interest for this scientific endeavor, you have been doing a great job so far!
However, in order to reduce the exploitability of the feature, we, in collaboration with MMOS and the HPA, have decided to stop calculating Accuracy Rating for the community consensus. Instead, we will increase the rate of training examples you get, for which you get an immediate rating.
This is only the first step in our attempt to improve Project Discovery and it's data quality, as we might introduce further changes to our scoring procedure. This change will also significantly reduce the heavy load on the MMOS servers and should ensure a smoother experience for everyone.
Feel free to leave any feedback and suggestions here, and may the cytoplasm be with you.
- CCP Wonderboy |
|
|
CCP Wonderboy
C C P C C P Alliance
11
|
Posted - 2016.03.21 11:58:23 -
[2] - Quote
Yadaryon Vondawn wrote:Excellent change! Could you also take a look at the presentation of example/training samples? In the in-game channel we noticed that for some reason every single control image is now Nucleus or Nucleoli. This skews accuracy heavily because they are easily recognized. They seem to be mostly in the 300-500 range.
This is for MMOS to decide, i'll forward this to them to see if they can mix things up a little.
Yadaryon Vondawn wrote:This is also tied to another request which I am sure you already have on the table, remove the sample number in the lower right corner. Only show it after one presses submit because people now know that they are control images and will thus vote in a specific way.
This is already implemented and should roll out soon(tm).
|
|
|
CCP Wonderboy
C C P C C P Alliance
22
|
Posted - 2016.03.23 13:55:06 -
[3] - Quote
ACESsiggy wrote:Be cool if the images loaded without having to wait 1 minute and the transmission errors for submission. Hopefully this update addressed these problems as well.
We should see significant performance increases with this next release. :) |
|
|
|
|