I like the COVID-19 project, but I find myself rather often in doubt about the graphs.
Is there anyone who created some info on what is what, suggestions for better demarcation etc?
@Aruar might be able to help.
He posted this thread stating he’s at Level 200 with 99% accuracy:
Thanks! I might do that indeed, but I’m also interested in the data behind the graphs.
And level 200… I’m only 50 at 99% yet.
I can help too. I’m at 252 with 99% accuracy.
I hope this helps:
You’re spending way too much time on yours. You’re making your bounds way tighter than they have to be to get high accuracy or would be beneficial to scientists. Reference my above post, cut your vertex count in half, you’ll increase your throughput accordingly.
With all do respect, just because they make videos or streams or talk about it or whatever doesn’t mean they’re the best. I’m watching their video and I go through samples faster than they do, and it’s not because I click faster, but because I click less. I do not know this person, so I am not making any commentary about them or their character, but as a general statement: anyone who is half as smart as they think they are knows there is always room for improvement, and even if they are in fact the very best, if there is something they can learn from someone of “lesser” talent, they should have the humility to adopt it.
Drawing 2 large boxes doesn’t help the scientists very much, even if the game gives you high accuracy points for it. A lot of those slides have multiple cells in the sample.
>>>IF<<< the quote from the scientist who helped CCP developed PD is true as cited in my post above (I have no reason to believe it is not), then they don’t care about the bounds, only the segmentation. This would be the disconnect between what CCP/players incorrectly think is good (a grossly excessive and unnecessary degree of granularity) vs what scientists actually need (quantity of cells and partitions between them). As such, there should be no guilt on anyone’s conscience regarding speed running PD via “hacky” methods since you are in fact helping scientists by using the crude boxing method as long as you box the correct cells (minor boundary overlaps or failure to capture all of cell are fine as long as the bulk of the cell is unambiguously captured)
Ooooookay. I’ll feel guilty about something else, then.
I must say that appreciate all the feedback here, and yes: i definitely did feel guilty about incorrect demarcations, even though I do try to be precise - and I feel that the known/verification samples nudge you into being extremely accurate (it’s also a fun challenge, but that’s beside my point).
On the other hand, I can also imagine that the scientists involved are also looking for training data for the AI, and that we mere humans are providing just that. Either way, thanks all