algorithmic modeling for Rhino
This file shows the basic idea of 2D sorting. If you don't understand why it requires an expression like:
from + (along * 1000)
Then read the following explanation from David Rutten which is part of this discussion
Ok, so here's the basic idea. We need to sort our rectangles according to two constraints (along and from). But sorting only works on a single list of numbers. This means we need to find a way to encode those two numbers into a single sortable list.
Imagine the following setup, three shapes (A, B & C) and a sorting guide:
If we only cared about sorting the shapes along the line, all we had to do was compose a list of all the ts (the parameters along the sorting guide closest to the shapes) and sort those. tA is smaller than tB and both are smaller than tC, so the order we'd get in this case would be A-B-C.
If we had two objects that had the same t value, but different distances from the sorting guide, there's no telling in what order they'll end up. It really is random due to the nature of the QuickSort implementation in .NET
If on the other hand we only cared about sorting the shapes based on distance from the guide, then we'd need to make a list of all the ds (the distances from the shape to the line) and sort those. dC is smaller than dAand both are smaller than dB, so the order in this case would be C-A-B.
But we want both. And we want the ts to be dominant over the ds. Meaning we want to sort all rectangles first along the guide, and then sort the ones in the same 'row' by distance from the guide.
We somehow need to combine t and d to give us a single numeric value that can be used to sort all the shapes in one go. Just adding them together won't work. After all, it's easy to imagine that tB + dB is about the same value as tC + dC.
Let's assign some actual values to the ts and ds and see how they behave:
If we were to simply add the t and d of each shape we'd get the following, sorting keys:
A = 5.0
B = 7.3
C = 5.0
D = 4.0
E = 8.8
F = 6.8
G = 7.2
H = 6.7
Which would result in the sequence {D, A?C, H, F, G, B, E}.
Since we want the ts to be dominant we need to amplify them so much that the biggest change in d cannot overrule the smallest change in t. I chose to multiply by a 1000, but you may need to experiment depending on specific case what sort of number works best. So let's include our amplification factor of 1000.0 for the ts and recompute the sorting keys:
A = 4001.0
B = 4802.5
C = 2502.5
D = 2501.5
E = 4804.0
F = 4802.0
G = 4003.2
H = 2504.2
now if we sort this, we get the proper sequence {D, C, H, A, G, F, B, E}
This is not the smartest of all 2D sorting algorithms btw. Since t is amplified so much it might wreck havoc with the sorting if there's a tiny amount of noise in the t values. For example if the rectangles aren't exactly in a grid, or maybe if the guide wasn't precisely parallel to the grid. However making a better sorter would require a fair amount of logic with probably a lot of if...then statements, so it's hard to do that in Grasshopper at present.
--
David Rutten
david@mcneel.com
Poprad, Slovakia
Tags:
Is the lack of pure, logical functionality in GH another limiting factor (see related post) in regards to its use for design realisation? I suppose the counter-question is, if logical operators were created as a set of components, would users possess the cognitive ability to successfully implement their use, or even care in the first place? I say this as using such methods will firstly require an acquired knowledge that adds additional complexity to a process, and secondly, they serve abstractly by nature which many individuals will fail to comprehend. This scenario would put in motion the possibility that such components would be consigned to the rubbish-heap due to user inability or complacency. Subsequently, the unique benefits logical operators / conditional statements offer as part of the design exploration process will be completely overlooked.
That is of course if it is possible to create such components in the first place. It could be said that the application of logical functions will always be problem-specific, meaning no pre-configured component will ever be adaptable enough to fully - or successfully - parse the infinite variation and structure of possible data input, and their pursuit as a workable addition to the GH repertoire is therefore futile, or simply impossible.
http://www.grasshopper3d.com/forum/topics/constrain-rotation?xg_sou...
If you have inquiries about grasshopper please post it here in the correct forum. This is a FAQ forum to help others, not to sell products :) Thanks
I am construing the point you made at the end of your post, hence the relevance. I did read the blog post you linked to, the day it was published in fact, and while i agree with some of the arguments put forward, i do feel some of the more important issues which i raised were not properly addressed.
I find with the way multi threaded discussions go things can get lost so I wasn't sure if you had seen it.
I don't think your post here is relevant to a common question asked by many about 2D sorting specifically. If you wish to add to the 2D sorting problem please do, but it should be restricted to helping or improving the methods for sorting in 2D in Grasshopper.
If you would like to start a Discussion in the Discussion Forum not the FAQs about this topic then your questions can be addressed there as the content here will be removed to keep it simple.
To develop this subject further, let me introduce polar coordinate system* sorting.
If this explanation is not clear, read it with attachment file opened.
*read about polar coordinate system here
As you may know, PCS (from now I will call polar coordinate system with PCS, and cartesian one with CCS) describes point position with 2 values (like x and y in CCS) which are r and theta(r,theta). r is for distance from PCS center, theta is angular dimension which is in 0 to 360 or 0 to 2*pi domain.
To hark back to David's guide line - here it is replaced with guide circle.
Why to sort points like this ? As usual, one image tells more...
Here is logic behind all this stuff :
how about non-planar curves ?
Any curve. It finds the closest parameter on the curve, then sorts by ascending parameter value.
--
David Rutten
david@mcneel.com
Poprad, Slovakia
Welcome to
Grasshopper
Added by Parametric House 0 Comments 0 Likes
Added by Parametric House 0 Comments 0 Likes
Added by Parametric House 0 Comments 0 Likes
Added by Parametric House 0 Comments 0 Likes
Added by Parametric House 0 Comments 0 Likes
© 2024 Created by Scott Davidson. Powered by