Dev-iL

@NickAlexeev Ah, I haven't considered that site. It does seem like a better fit there. Thanks!

Hi everybody! Do you think that this question would be {more} on-topic on your site {than on SO}?

Not even "formally" so much as "clearly enough". I'll know I'm on the right path when I can at least explain my problem in a non-ambiguous manner to the people in this channel :)

@whuber Thanks for your comment! One can think of this as a parameter estimation problem, where the optimal solution minimized the SSE of the system of equations. To be more specific, I have 3 unknown and M measurements (equations), and I'm looking for ways to justify that M of 3 or 4 is sufficient. As mentioned, the data is practically noiseless, but there is some unknown degree of correlation between the measurements. One of my issues is that I'm not sure how to state this problem formally.

... so if the "unique" part is sufficiently dominant I can reach a useful solution, but otherwise I cannot

Hi everybody! I need some help putting thoughts into words.. I'm trying to express the idea that finding an optimal solution to a system of N equations requires at least as many measurements, but also that these measurements need to be "sufficiently uncorrelated/independent" (because if they're "too correlated" I would need some M>N measurements to have enough "information" to reach an optimum). Measurements are not random, but rather something like v[1,2...] = v_common + v_unique[1,2,...] ...

@gung Thank you for your comment on the question - I'm happy to hear that it's suitable for the site :)

I've posted this as a question now.

(Unless what constitutes "off topic" on Cross Validated doesn't cover this case)

@gung Normally, I would do that. However, what would be the point of asking "formally" if its going to be closed as off topic?

... fast knn-search algorithm to speed up the leave-one-out error computation. Is there something "costly" I can compute once (kdtree?) and use it compute the error for all the points? The desired outcome in my case is a fast evaluation of the xval error (to compare how bad is the "folding" effect among hundreds of these hypersurfaces). Any other algorithm that could quantify the folding effect would also work. Thanks for any help! [2/2]

[1/2] Hi guys, I've been hoping you could point me in the right direction (I suppose it's a bit of a "tool request"): I have a dataset of ~7k scattered points in 3D which represents a hypersurface that may or may not "fold unto itself". I know that I can detect if such folding occurs by examining the errors when performing leave-1-out cross-validation. I also know that CUDA-based algorithms exist to perform knn search. However, as I don't come from this field I can't tell if I can leverage a ...

Hi guys! I have a math related question which seems to be too broad for math.se, so I hope chat can be a suitable alternative.. In short, I'm looking for ideas regarding a suitable fitness function(s) for optimization and could really use the insights of somebody with a better understanding of spans, spaces etc. I have already explained it in "my home chatroom" (I don't want to spam this room seeing how I'm a newcomer...)