Very small sections of neuron...
  • Well, since the server is down I might as well test out the communications apparatus.

    When I started a few days ago it seemed that the system would only give me very small chunks to check. I mean like a little dot in the corner or a sliver on an edge. Point is that (my assumption) if you don't actually color anything in there is no credit given. Now I get that this is not a credit sort of system, as in the credits aren't worth anything but bragging rights. But the thing is that at first I thought something was wrong, I hadn't done the work properly or something like that. When I finally did get a chunk that had work to be done I did get a few points. Then it was right back to the slivers and dots. I aborted several units and tried to describe as best I could about why. Then today it actually gave me a decent chunk to work with, there was plenty of holes to fill and everything went much as I had expected it to. So I guess for the future users and myself can I get some confirmation that the system will only rank your task if there was some holes to fill in(computer mistakes / omissions)?

    Also, it may not be a bad idea to give a point or two for the sliver and dot units so that people don't get discouraged or think that they are doing something wrong. In any case this is fun when it works, can't wait to get back into it when the servers are back online. Thanks to everybody who thought this up and made it possible. Good luck and happy connectome-ing.

    Turing
  • Hey Turing,

    Thanks for the question.  You've honed right in on the 2 big things we are working on at the moment.  The first is the task generation.  We don't want to be giving you guys all the little worthless tasks (both in points for you and in science).  Look forward to never seeing another small chunk again in the next week or two.

    The second thing that we are working on is the scoring system.  We recognize that it can be a bit demoralizing to get 0 points for a task.  At the same time, we can't just give out points for everything because we don't want people to get points for skipping tasks.  It's a difficult problem because we don't actually know what the answer is beforehand, so sometimes it's tough to tell whether you made a mistake or the people who came before you.  In any case, we're open to suggestions and we'll try to come out with a better scoring system in the next few weeks.

    Matt
  • Hi Matt, I think that it wood be nice to have a number, telling you how many tasks you have completet too. I mean, it can take just as long to see all the slices in a task too tjek that everything is allready ok, as it takes to fill in a frew missing spots in a nother task. My geuss is that both tasks is importent.

    Jens

    PS. I think I have done some of the tasks, more than once. whay is that?
  • Jens, 

    We have been discussing this problem for a while now, and are trying to figure out a solution. I agree with you that we should reward people a little when they check the small chunks. 

    I'm not sure why the same task would show up again. Did this happen in the same session or in different sessions?

    Sebastian
  • Hi Sebastian, it has been in fifferent sessions, before and after your server breakdown a frew days ago.

    About the point, I think that it would be nice to have the point score and the number of tasks completed at the same time.
    Its not nesseserely at small chunk, that give 0 point. It can also be a big one, but the computer have drawn it perfect and there is nothing to add = 0 point, but it still takes a lot of time to check it in details.

    Jens
  • PS. The "sign out" dont seem to work for me.

    Jens
  • Hello everybody,

    I really like the improvements that have been made. Far fewer dots and slivers and the control options are better. I also would like to second some of the ideas here about scoring mechanisms. It does take time to just go through the motions. Perhaps a score of .01 per slice viewed or something like that (any chance we can see the algorithm being used to rank user's work?) It's tedious enough that I don't think you would have (m)any people just scrolling through for the points; there's a certain dedication to a task like this right.

    I also like the idea of a progress counter, how many slides viewed, points earned for current unit, total points to date, total units finished, etc. ect.

    That said... Today I did a few units and encountered a new beast. Several (including a huge ?node, cell body? that occupied about a third of the volume of the unit and only the upper right quadrant available in 3D view, I'll get back to this one) that when I clicked "I'm Finished" reported that I was "the trailblazer" (that's cool) but it seems no points were awarded for the units. Even more curious, the huge chunk I mentioned had a ton of holes, it spanned the whole unit's worth of slides and seemed to correctly color in the blobs (not the whole form but all of the bits that at a good spacial guess fit in the huge node in the 3D view that was only partially rendered. This unit took about half an hour, maybe more, and then the "trailblazer" message and no points. So something is still off, maybe, I don't know, just typing out loud.

    Cheers,
    Turing
  • Hi turing8, 

    Regarding the new beast you encountered ... You are right, it's a part of a cell body and we intended to exclude any task containing cell bodies because of the exact issues you mentioned. Apparently a bad guy has escaped our constraining. We will look into it and won't let the cell body tasks out again. For the moment, just abort such tasks. Sorry for the inconvenience!

    JK
  • I've seen a few of these massive zones as well today (16th march). One gave some points, but I encountered the same problem of spending ages correcting the automated algorithm for zero points. Also, (working in firefox), the updating on these large tasks is extremely slow and seems to not update the 3d view with anything intelligible, even after completing the task - actually chrome does the same thing, but is faster with the new code update.

    I would insert a picture, but it only allows http links.

    I'm guessing its so slow because its operating using javascript in a browser.

    Excellent project by the way - I'm more hooked on this than foldit :)

    Edit - some of the performance problems are stemming from the 3D widget; killing that speeds it up enormously.



  • A few more words of explanation for the curious...you color neurons on EyeWire by guiding an artificial intelligence (AI). The AI was trained to color the branches of neurons. It was not trained to color cell bodies, which is why it does badly with them. We have tried to exclude cell bodies from EyeWire tasks, but occasionally one of them slips through and becomes an annoyance. Feel free to abort such tasks.

    To solve this problem completely, we are planning to improve the AI to color the cell bodies accurately.
  • It really wasn't annoying. It was actually kind of cool because there was so much to do. The only annoying part was that I spent so much time on it and got a digital doughnut. But I suppose there is a better solution than having the humans do it manually.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

In this Discussion