[erlang-questions] Trying to learn the Erlang Way

kraythe . <>
Wed Feb 12 18:56:06 CET 2014


Ahh octtree is good if the objects are static and if your goal is to find
an object in a large space. It is a reductive search algorithm. In this
case all of the objects are in motion and I want to know my neigbors, not
find object X in the space. Therefore a different approach is called for.

 Imagine thousands of particles flowing around in a medium and needing to
determine their interactions based on vicinity, momentum, etc. From frame
to frame the state of your tree would change and the cost would make the
simulation impossible to render in realtime. Instead if the world is
divided up into cubes that are larger than the sphere of influence and each
cube knows about its neighbors then I can easily get a candidate list as
long as the particles update the cube entity that they are in. As they move
into a cube, they inform the cube that they have entered and as they leave
the cube, they inform it of their exit. If I know the cube I am in, and I
know my sphere of influence overlaps into another cube, I can immediately
exclude all candidate not in either cube. If the cube has references to its
neigbors, the solution is easy.

1) Determine if I overlap sphere into neighbors (simple math problem of the
radius of my sphere and the boinds of the cube.
2) For all cubes I overlap, get members.
3) For all members determine which are inside my sphere.
4) Interract with elements in my sphere.

Incidentally my current version of the code, Thanks to Richard's input,
 is:

%% Subtract the second vector from the first
subtract({X1, Y1, Z1}, {X2, Y2, Z2}) -> {(X1 - X2), (Y1 - Y2), (Z1 - Z2)}.

%% Compute if the vector T is in the sphere with center at vector C and
with radius R.
in_sphere(C, R, T) ->
  {Dx, Dy, Dz} = subtract(C, T),
  Dx * Dx + Dy * Dy + Dz * Dz =< R * R.

%% Calculate all entities in the sphere defined at center vector C with
radius R and using Fpos which will extract the location to check from
%% each entity in the list. The method Fpos must return a three tuple
vector and take the entity as an argument.
entities_in_sphere(C ,R, Entities, Fpos) ->
  [E || E <- Entities, in_sphere(C, R, Fpos(E))].

It also appears to run quite fast according to preliminary timer:tc tests
but of course there is need for more massive testing of load. I may still
have to use the type optimization richard pointed out. I am debating
staying with integer arithmetic also.


*Robert Simmons Jr. MSc. *


On Wed, Feb 12, 2014 at 8:04 AM, Joe Armstrong <> wrote:

>
>
>
> On Tue, Feb 11, 2014 at 11:17 PM, kraythe . <> wrote:
>
>> Richard,
>>
>> I appreciate your response and the effort you put into it. And I learned
>> a lot from it. In this case I am learning the thinking mode of Erlang as it
>> is different a bit from what I do to pay the bills. I have yet to get into
>> list comprehensions in Erlang so that is on my ... well ... list. :) You
>> have provided valuable insight.
>>
>> The main focus of the method is if I have a number of objects with a
>> vector position in a simulation and I want to exclude considerations of
>> interactions with objects outside the sphere of influence, I have to
>> quickly discard the candidates that are not inside the sphere of influence.
>> I originally thought to write a method that did just the vector math
>> because I was wondering if that kind of math would have to ultimately be
>> turned into something native. Even a delay of 1 second would be fatal to
>> the simulation.
>>
>
> Isn't this what octrees are for? - You'd have to use a "cube of influence"
> - but as far as I know octrees
> can be used to rapidly partition objects in a 3-d space
>
> Take a look at this http://en.wikipedia.org/wiki/Octree
>
> /Joe
>
>
>>
>> The algorithm, however, shouldnt have to consider all candidates as the
>> world is broken into spacial segments (cubes) such that the sphere could
>> intersect with at maximum 8 neighboring cubes so I would only need to
>> consider simulation objects within those 8 cube spaces when determining
>> which elements were actually within the sphere of influence. I have been
>> trying to devise a method of reducing the candidate set even further and am
>> still working on that. Perhaps edge detection and cube boundary
>> calculations but I don't want to spend more math doing that then I would
>> simply excluding objects vector by vector.
>>
>> Anyway thanks for the reply, definitely some information that I can use
>> in there.
>>
>> *Robert Simmons Jr. MSc.*
>>
>> On Sun, Feb 9, 2014 at 10:59 PM, Richard A. O'Keefe <>wrote:
>>
>>>
>>> On 8/02/2014, at 4:53 AM, kraythe . wrote:
>>> > Anyway back to the subject at hand. The algorithm is set but now I am
>>> at another quandary Lets say these vectors represent a position in space of
>>> particular objects in a simulation. The process of culling the vectors
>>> based on the sphere is entirely a vector problem but what the user calling
>>> cull/3 really needs to know is which objects are not culled from the list,
>>> not just which vectors are not culled. Now in Java I could do a number of
>>> things if I wanted to keep the cull algorithm as it is. I could return the
>>> list of integers containing the original indexes of the vectors in the list
>>> that were culled and use that to filter out which objects need to be
>>> considered for the simulation step.
>>>
>>> The word "cull" really grates.
>>>
>>> And all those negations *really* confused me *all over again*.
>>> I wrote a lengthy and helpful description of how to get the
>>> points that were accepted and the points that were rejected
>>> as two lists, because that was the only way I could interpret
>>> your question to make sense in Erlang.  But on repeated re-reading
>>> it became clear that you were asking for something else.
>>>
>>> There is no such thing in Erlang as object identity.
>>> The distinction you are drawing between the "points" and the
>>> "objects" simply doesn't exist.
>>>
>>> It so happens that if you use a list comprehension like
>>>         [P || P <- Points, some_predicate(P)]
>>> the elements of the result *will* be (references to) the same
>>> implementation-level webs of chunks of memory that were in the
>>> original list, not copies.  But nothing other than performance
>>> depends on this.
>>>
>>> A list of integers such as you ask for could be obtained, but it
>>> would be very little use to you, because Erlang lists are *LINKED
>>> LISTS*, not *INDEXED ARRAYS*.   Finding the nth element of a list
>>> takes O(n) time, and that could not be changed without making
>>> lists much less useful.
>>>
>>>
>>>
>>
>> _______________________________________________
>> erlang-questions mailing list
>> 
>> http://erlang.org/mailman/listinfo/erlang-questions
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20140212/d84559d2/attachment.html>


More information about the erlang-questions mailing list