Dinkydau
|
|
« on: December 23, 2016, 09:08:08 PM » |
|
This must be the best breakthrough since perturbation! I just tested it a few times and it really works. It's simple and elegant. This makes zooming so much easier.
Around 2^1000 it starts to become much slower. I remember that being a depth where other things get slower as well. I would guess it's about as fast as manual zooming from there on, except you can do other things while kalles fraktaler is busy.
|
|
« Last Edit: December 24, 2016, 01:12:20 PM by Dinkydau »
|
Logged
|
|
|
|
Kalles Fraktaler
|
|
« Reply #1 on: December 28, 2016, 12:19:57 AM » |
|
This must be the best breakthrough since perturbation! I just tested it a few times and it really works. It's simple and elegant. This makes zooming so much easier.
Around 2^1000 it starts to become much slower. I remember that being a depth where other things get slower as well. I would guess it's about as fast as manual zooming from there on, except you can do other things while kalles fraktaler is busy.
Yes it is indeed a huge breakthrough for me at least. And I think it is much faster than manual zooming all the way down. It took 2 months for me to reach e10000 the first time with manual and automatic zoom, but only some week to reach e20000 with NR... If you use zoom size 32, and zoom to e20000 manually, you would need +6600 clicks... All thanks to the generously shared code by claude!
|
|
« Last Edit: December 28, 2016, 12:46:05 AM by Kalles Fraktaler, Reason: Rephrase »
|
Logged
|
|
|
|
Adam Majewski
|
|
« Reply #2 on: December 28, 2016, 01:01:39 PM » |
|
Where is more description / code ?
|
|
|
Logged
|
|
|
|
|
Kalles Fraktaler
|
|
« Reply #4 on: December 28, 2016, 10:36:14 PM » |
|
I convert the zoom of the final minibrot to powers of 2. Then multiply it with 3/4 and convert back to zoom. It works well for most cases. Some times a few zoom-in or -out is needed, it is easy to determine by the density of the resulting view. However when making short-cuts, as described by TheRedShifter, this doesn't work so well
|
|
|
Logged
|
|
|
|
Adam Majewski
|
|
« Reply #5 on: December 31, 2016, 02:04:10 PM » |
|
i understand idea of perturbation method. Why this method is better then perturbatiion ? Can you describe it in simple words ?
|
|
|
Logged
|
|
|
|
quaz0r
Fractal Molossus
Posts: 652
|
|
« Reply #6 on: December 31, 2016, 03:34:34 PM » |
|
it is not a different rendering method (they are still using perturbation). it is automated zooming. they use newton's method to find and progressively refine the precision of the location of the minibrot at the center of a pattern. this allows them to arbitrarily select a magnification between the location they started at and the final minibrot they calculate to be at the center of that location.
|
|
|
Logged
|
|
|
|
Dinkydau
|
|
« Reply #7 on: January 02, 2017, 08:17:07 AM » |
|
Thanks for the links, Claude. I was wondering how this worked and I couldn't find anything.
|
|
|
Logged
|
|
|
|
Dinkydau
|
|
« Reply #8 on: January 02, 2017, 08:43:14 AM » |
|
Yes it is indeed a huge breakthrough for me at least. And I think it is much faster than manual zooming all the way down. It took 2 months for me to reach e10000 the first time with manual and automatic zoom, but only some week to reach e20000 with NR... If you use zoom size 32, and zoom to e20000 manually, you would need +6600 clicks... All thanks to the generously shared code by claude! It must help that your zooms to depths like E20000 were low on iterations. That makes the process of finding the period a lot shorter. I'm currently using it to zoom to E2538 and I can see this making faster progress than I could have by manual zooming in mandel machine (~ 2^6,3 magnification per click, very efficient reference point calculation). I estimate the period is roughly 87 million. Reference points would take close to an hour eventually. It's probably just right after the slowdown around 2^1000 that manual zooming can be faster.
|
|
« Last Edit: January 02, 2017, 08:49:23 AM by Dinkydau »
|
Logged
|
|
|
|
Kalles Fraktaler
|
|
« Reply #9 on: January 02, 2017, 02:08:01 PM » |
|
It must help that your zooms to depths like E20000 were low on iterations. That makes the process of finding the period a lot shorter. I'm currently using it to zoom to E2538 and I can see this making faster progress than I could have by manual zooming in mandel machine (~ 2^6,3 magnification per click, very efficient reference point calculation). I estimate the period is roughly 87 million. Reference points would take close to an hour eventually. It's probably just right after the slowdown around 2^1000 that manual zooming can be faster.
Yes you are right, the amount of iterations and period number has a big impact on the speed of NR calculations, since every loop is iterating the iterations. And every shape stacking is doubling the iterations/period, so the number increases quickly and exponential. So the depth itself doesn't have much impact
|
|
|
Logged
|
|
|
|
Dinkydau
|
|
« Reply #10 on: January 06, 2017, 10:40:39 AM » |
|
A negative percentage Let's hope this is not a bad sign.
|
|
|
Logged
|
|
|
|
Kalles Fraktaler
|
|
« Reply #11 on: February 16, 2017, 09:43:56 PM » |
|
A negative percentage Let's hope this is not a bad sign.
<Quoted Image Removed>
No it will work 100*progress/max where 100*progress overflows maximum integer value of 2147483648 So you have a period of more than 21,474,836? This occurs also when calculcating the reference if you have more than 21,474,836 iterations...
|
|
|
Logged
|
|
|
|
Dinkydau
|
|
« Reply #12 on: February 16, 2017, 11:39:37 PM » |
|
That explains it. I've seen it happen a lot with the reference but no errors ever occurred because of it.
It's still not done because I assigned the process a low prority and I'm using the Newton-Raphson zooming very intensively. I've got 9 instances of Kalles Fraktaler open at the moment.
|
|
|
Logged
|
|
|
|
|