<div dir="auto">Nice graphics!</div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri, Sep 1, 2023, 4:25 PM Kristofer Munsterhjelm <<a href="mailto:km_elmet@t-online.de">km_elmet@t-online.de</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">On a completely different subject, I've added quasi-Monte Carlo support <br>
to quadelect's Yee diagram drawing module, and it seems to be making <br>
quite a difference.<br>
<br>
Here's an example with a particular nonmontone method (not recommended <br>
for practical use).<br>
<br>
Ordinary MC, 2000 voters: <br>
<a href="https://munsterhjelm.no/km/elections/quasi_monte_carlo_yee/monte_carlo_2000.png" rel="noreferrer noreferrer" target="_blank">https://munsterhjelm.no/km/elections/quasi_monte_carlo_yee/monte_carlo_2000.png</a><br>
<br>
QMC, 200 voters: <br>
<a href="https://munsterhjelm.no/km/elections/quasi_monte_carlo_yee/quasi_mc_200.png" rel="noreferrer noreferrer" target="_blank">https://munsterhjelm.no/km/elections/quasi_monte_carlo_yee/quasi_mc_200.png</a><br>
<br>
QMC, 2000 voters: <br>
<a href="https://munsterhjelm.no/km/elections/quasi_monte_carlo_yee/quasi_mc_2000.png" rel="noreferrer noreferrer" target="_blank">https://munsterhjelm.no/km/elections/quasi_monte_carlo_yee/quasi_mc_2000.png</a><br>
<br>
QMC, 8000 voters: <br>
<a href="https://munsterhjelm.no/km/elections/quasi_monte_carlo_yee/quasi_mc_8000.png" rel="noreferrer noreferrer" target="_blank">https://munsterhjelm.no/km/elections/quasi_monte_carlo_yee/quasi_mc_8000.png</a><br>
<br>
The 200 voters QMC picture looks roughly similar to the 2000 voters MC <br>
one. "Just" a 10x improvement doesn't sound like much, but if QMC <br>
converges as 1/n and MC at 1/sqrt(n), then it'll only get better at <br>
higher vote counts.<br>
<br>
There's another oops: it turns out Warren implemented something kind of <br>
similar to a low discrepancy Gaussian sampling for his Yee code. <br>
(Basically he picks regularly spaced angles around a circle, and <br>
distance from center randomly.) So that explains why IEVS was getting so <br>
much sharper images than say, Brian Olson's code. But IEVS is limited to <br>
200x200, and it more or less tops out at ~10k voters due to his <br>
progressive sampling code; any noise still present at 10k voters remains <br>
even at 65k.<br>
<br>
Quadelect is still 11x slower than IEVS per voter, so there's definitely <br>
room for more general optimization. But having implemented QMC should <br>
make it easier to adapt it to other domains like strategy resistance <br>
calculations.<br>
<br>
-km<br>
----<br>
Election-Methods mailing list - see <a href="https://electorama.com/em" rel="noreferrer noreferrer" target="_blank">https://electorama.com/em</a> for list info<br>
</blockquote></div>