The good news and the bad news is that we got a little sidetracked into the area of social media, launchingProgress Thru Processorson Facebook. And this has taken up a lot of our development bandwidth. As hasCharity Engine.
At GR, the next significant development milestone is a long-in-the-works update of the site infrastructure -- we're converting everything to Drupal. (*Yes, this has been the aspiration for a silly long time: but should be soon: seriously this time.) This improved instrstructure will make lots of stuff easier to impliment. But even after that -- misc competitive points schemes are going to take a bit, given our resources. (*Anyone want to find a grant for this purpse?)
I wonder, after you installed -- In your desktop software (the BOINC software, particularly), did you use the tools > attach to account manager function?...
* One last thing -- small point -- the result you got on clicking the "comment" button is in fact just an awkward way of saying that page is at present empty. You could put stuff in there if you wanted; but i wouldn't bother -- i think discussions are a better format for comments....
Aaron -- seems our posts were more or less simultaneous; let me know if the above gives you better insight into what we're trying to accomplish.
But to describe in one other way:
We aim to create a set of computer games. And the idea of the games is to present information to users sufficient for them to make a good "guess" about optimal model parameters. And to your point above, the thought is that people will be able to do this better than an evolutionary algorithm alone.
(*In large part we think people will do it better because we can't afford to let the evolutionary algorthm churn though billions of iterations-- that's too expensive, even taking into account the fact that there are tens of thousands of PCs contributing to the effort.... So the objective is to bring some human intelligence into the mix, to guide the process a bit...)
Returning to the original question: we want to produce a range of visualizations; and then we want to create associated tools for the user to predict where next to search next....
But here is the second layer of complication: there are not one or two or three or five parameters in our dataset to present to the user. There are 25. And we can't think, for now, of how to present a 25-dimensional dataset to a person.
And so the thought is to see if we (and by this I mean the "you" of "we") can develop methods to reduce the number of parameters presented to the user.
Think of a machine that has 25 levers (the parameters). What we need is an "interface" of some 5 levers or so: the user works with these 5, which are linked to the underlying 25 in some ordered way; ie when the user sets the 5 levers, this input can be meaningfully passed ("translated") to the underlying machine's 25 levers…
Not sure if that's the best description; but maybe it helps?
Yes, I think your description of the project is close… You write:
"We're just trying to visualize the logs of an evolutionary model optimizer? Why? Is there real hope that you can make a better model by presenting a bunch of unlabeled parameters and looking for input from "human intuition"?
So: if we had a handful of parameters, 3 or 5, then your description would be spot on: we'd just want to visualize the logs; and yes, the thinking is that a person, on looking at such a presentation, could see trends; and so a person could "jump ahead" of the evolutionary algorithm, which plod along in very small steps.
By way of imagining what I mean: let's say the results could be plotted on a 2D chart… a thousand initial runs would produce a fairly random scatter… subsequent runs would begin to cluster; you'd see small circles of growing density, where the evolutionary optimizer would run subsequent generations in the vicinity of those that had worked in the past.
So the idea is that a person could see such general patters emerging; ie the circle of results forming; and a person could then "stick a pin" in the middle of that circle, on the basis of the reasonable assumption that that's where things are going. And many times, this will save a number of generations of iterations. And since in our case model runs are computationally expensive, this would be a significant result.
There is a pretty "normal" (i think) wiki -- an overview of the project ishere. From that page there is a link to another wiki page, wherein I was hoping maybe people would help build out adescription of alternative approaches. (ie, maybe people who don't have time to code, might at least comment on general directions.)
Sorry -- unfortunately, via GR, there is not a way to have separate resource shares for individual computers. This is the sort of design tradeoff one makes to keep the user interface simple.
(*You could create two accounts -- one for each computer. But that's not a great solution, either...)
...
Page 2 of 23
Loading...
Please enable JavaScript support in Your web browser to display this page property.
