to incorporating math and geometry in computational design education, Paneling Tools
Marlo Ransdell, PhD Creative Director, at FSU , Digital Fabrication in Design Research and Education
Andy Payne, LIFT architects | Harvard GSD | FireFly
Jay H Song, Chair, Jewelry School of Design, Jewelry as Personal Expression, Extra+Ordinary@Jewelry.com
Pei- Jung (P.J.) Chen, Professor of Jewelry, SCAD
Gustavo Fontana, designer/co-founder nimbistand, Diseñar, desarrollar y comercializar productos por tu cuenta.
Joe Anand, CEO MecSoft Corporation, RhinoCAM
Julian Ossa, Chair, Industrial Design Director, Diseño – Una opción de vida a todo vapor!, UPB
Minche Mena, SHINE Architecture, Principal
J. Alstan Jakubiec, Daylighting and Environmental Performance in Architectural Design Solemma, LLC
Carlos Garnier R&D Director / Jaime Cadena – General Director, Plug Design, www.plugdesign.com.mx
Mario Nakov, www.chaosgroup.com [ V-Ray ]
Andres Gonzalez, RhinoFabStudio
Workshops:
o) Paneling Tools
o) RhinoCAM
o) Rhinology in Design, for Jewelry
o) Footwear
o) V-Ray: Jewelry Design
o) V-Ray: Architects and Industrial Designers
o) FireFly
o) J. Alstan Jakubiec, DIVA
The cost for each workshop or the Lectures is 95.0 US$
To register:
WORK-SHOPS April 2 - RHINO DAY
WORK-SHOPS April 3 - RHINO DAY
REGISTRATION RHINO DAY
NOTE: All students and faculty members that register to this event, will receive a Rhino 5 Educational License at the event.
…
- nickname is rather the best approach - and not on active group, but that's irrelevant anyway).
Step back (assuming that you are talking about the "Tens_from_random_blah_blah" definition):
1. Engineering is the art of demystifying (or we are promising that anyway, he he). This means that you start defining (better: outlining) some topology for things based on some "generic" rules (like the ones applied for the masts,cables,cones etc etc). These things are kept in some kind of structure (Lists, DataTrees etc). Things are few in 99.99999% of cases (i.e. : even the biggest membrane "module" has, say, 20-50 masts per "module").
2. Then ... handling things "individually" (mostly modifying) becomes the most critical part. See this (an x "possible" solution by combining a myriad of "options" : a no cones membrane solution, in plain English):
3. But the above is impossible (for more than obvious reasons). You should deploy masts in some high/low sequence in order to achieve some meaningful convex/concave formation that could work.
4. This "works" : 5. This doesn't:
6. This works partially (the formation at the back is "flat" == undo able):
7. This is utterly kitsch (and faulty as the case6 - the back portion):
So it's quite obvious that without a (quite complex) capability to individually control things (in this occasion : mast heights) the whole definition is a waste of computer time. Additionally the more the solution is "demystified" (some curve is defined, some random points are created, some masts are in place, some cables appear etc etc) the more additional constrains are required in order to "narrow" the possibilities (In plain English : sliders should control other sliders as regards their min/max values, true/false, you/me etc etc).
Remember that we are talking about ONE (mast height) out of a myriad things that you should control "manually" (it's utterly pointless to mastermind some kind of "generic" rules - or use naive attractors etc etc) .You'll see the difference when I'll completely reform the definition by adding individual control upon anything.
PS: what about the blocks? (the real life stuff that actually make any solution possible). Can you imagine a 2nd set of "restrictions" imposed by "a child to his parent"? (Assembly/Component modeling , that is).
more soon
…
ly this is a Rhino.Python problem and not a Grasshopper issue, but it could apply to both!
I was trying to take a simple example of moving a ball around and see how it could be animated through Rhino.Python. The code works great in wire frame with now memory issues at all. However, when I switch the view to Shaded or Rendered, things go south pretty quickly. The RAM usage of Rhino which was steady around 350mb (ish) now grows every frame after a minute or so, it is in the GB's and never drops even after the script has stopped.What gives? Clearly this must be possible because Bongo does something similar when it does animations. Check out my code below and I would love to hear your thoughts.
import time
import rhinoscriptsyntax as rs
import Rhino
height = 100
width = 100
x = 0
y = 0
xspeed = .1
yspeed = .3
start_time = time.time()
end_time = 60
run_time = 0
sphere = rs.AddSphere((x,y,0), 5)
while run_time < end_time:
x = x + xspeed
y = y + yspeed
if x > width/2 or x < -width/2:
xspeed = xspeed * -1
if y > height/2 or y < -height/2:
yspeed = yspeed * -1
rs.MoveObject(sphere, (xspeed, yspeed, 0))
Rhino.RhinoApp.Wait()
run_time = time.time() - start_time…
uick answers. Below you will find some suggestions, but don't think of them as rules and especially don't think of them as guarantees.
1. Choose a descriptive title for your post
Don't call your question "Help!" or "I have a problem" or "Deadline tonight!", but actually describe the problem you are having.
2. Be succinct but clear in your wording
People need to know some details about your problem in order to understand what sort of answers would satisfy you, but nobody cares about how angry your boss or how bad your teacher or how tight your deadline is. Talk about the problem and only the problem. If you don't speak English well, you should probably post in your native language as well as providing a Google Translation of your question.
3. Attach minimal versions of all the relevant files
If you have a GH/GHX file you have a question about, attach it to the post. Don't expect that people will recreate a file based on a screen-shot because that's a lot of pointless work. It's also a good idea to remove everything non-essential from a GH file. You can use the 'Internalise Data' menu option to cut everything to the left of a parameter:
If you're importing curves or Breps or meshes from Rhino, you can also internalise them so you won't have to post a 3DM file as well as a GH file. If you do attach large files, consider zipping them first. Do not use RAR, Ning doesn't handle it.
It is especially a good idea to post files that don't require any non-standard components if at all possible. Not everyone has Kangaroo or Hoopsnake or Geco installed so if your file relies on those components, it might not open correctly elsewhere.
4. Include a detailed image of the GH file if it makes sense
If your question is about a specific (group of) components, consider adding a screenshot of the file in the text of the post. You can use the Ctrl+Shift+Q feature in Grasshopper to quickly create nice screenshots with focus rectangles such as this:
5. Include links to online resources if possible
If you have a question about Schwarz Minimal surfaces, please link to a website which talks about these.
6. Create new topics rather than continuing old ones
It's usually better to start a fresh question, even if there's already a discussion that kinda sorta tangentially touches upon the same issue. Please link to that discussion, but start anew.
7. This is not a 'do my work for me' group
Many of us like to help, but it's good to see effort on our part being matched by effort on your part. Questions in the form of 'I need to do X but cannot be bothered to try and learn the software' will (and should) go unanswered.
7b. Similarly, questions in the form of 'How do I quickly recreate this facade that took a team of skilled professionals four months to figure out?' have a very low success rate.
--
David Rutten
Lead Grasshopper Development
Robert McNeel & Associates…
Added by David Rutten at 12:58pm on October 1, 2013
ting.
Thanks
Rania
** Warning ** IP: Note -- Some missing fields have been filled with defaults. See the audit output file for details.
** Warning ** Version: in IDF="'8.2.7'" not the same as expected="8.2"
** Warning ** ManageSizing: For a zone sizing run, there must be at least 1 Sizing:Zone input object. SimulationControl Zone Sizing option ignored.
** Warning ** ManageSizing: For a plant sizing run, there must be at least 1 Sizing:Plant object input. SimulationControl Plant Sizing option ignored.
************* Testing Individual Branch Integrity
************* All Branches passed integrity testing
************* Testing Individual Supply Air Path Integrity
************* All Supply Air Paths passed integrity testing
************* Testing Individual Return Air Path Integrity
************* All Return Air Paths passed integrity testing
************* No node connection errors were found.
************* Beginning Simulation
************* Simulation Error Summary *************
** Warning ** The following Report Variables were requested but not generated
** ~~~ ** because IDF did not contain these elements or misspelled variable name -- check .rdd file
************* Key=*, VarName=ZONE IDEAL LOADS SUPPLY AIR TOTAL COOLING ENERGY, Frequency=Hourly
************* Key=*, VarName=ZONE IDEAL LOADS SUPPLY AIR TOTAL HEATING ENERGY, Frequency=Hourly
************* Key=*, VarName=ZONE PACKAGED TERMINAL HEAT PUMP TOTAL COOLING ENERGY, Frequency=Hourly
************* Key=*, VarName=ZONE PACKAGED TERMINAL HEAT PUMP TOTAL HEATING ENERGY, Frequency=Hourly
************* Key=*, VarName=CHILLER ELECTRIC ENERGY, Frequency=Hourly
************* Key=*, VarName=BOILER HEATING ENERGY, Frequency=Hourly
************* Key=*, VarName=FAN ELECTRIC ENERGY, Frequency=Hourly
************* Key=*, VarName=ZONE IDEAL LOADS SUPPLY AIR LATENT HEATING ENERGY, Frequency=Hourly
************* Key=*, VarName=ZONE IDEAL LOADS SUPPLY AIR LATENT COOLING ENERGY, Frequency=Hourly
************* Key=*, VarName=ZONE IDEAL LOADS SUPPLY AIR SENSIBLE HEATING ENERGY, Frequency=Hourly
************* Key=*, VarName=ZONE IDEAL LOADS SUPPLY AIR SENSIBLE COOLING ENERGY, Frequency=Hourly
************* Key=*, VarName=SYSTEM NODE MASS FLOW RATE, Frequency=Hourly
************* Key=*, VarName=SYSTEM NODE TEMPERATURE, Frequency=Hourly
************* Key=*, VarName=SYSTEM NODE RELATIVE HUMIDITY, Frequency=Hourly
************* There are 3 unused schedules in input.
************* There are 5 unused week schedules in input.
************* There are 13 unused day schedules in input.
************* Use Output:Diagnostics,DisplayUnusedSchedules; to see them.
*************
************* ===== Recurring Surface Error Summary =====
************* The following surface error messages occurred.
*************
************* Base Surface does not surround subsurface errors occuring...
************* Check that the GlobalGeometryRules object is expressing the proper starting corner and direction [CounterClockwise/Clockwise]
*************
** Warning ** Base surface does not surround subsurface (CHKSBS), Overlap Status=No-Overlap
** ~~~ ** The base surround errors occurred 1 times.
** ~~~ ** Surface "839A5ADACCE44BC0AF00_GLZP_31" misses SubSurface "839A5ADACCE44BC0AF00_GLZP_31_GLZ_31"
** Warning ** Base surface does not surround subsurface (CHKSBS), Overlap Status=Partial-Overlap
** ~~~ ** The base surround errors occurred 1 times.
** ~~~ ** Surface "839A5ADACCE44BC0AF00_GLZP_34" overlaps SubSurface "839A5ADACCE44BC0AF00_GLZP_34_GLZ_34"
*************
** ~~~ ** The base surround errors occurred 2 times (total).
*************
************* EnergyPlus Warmup Error Summary. During Warmup: 0 Warning; 0 Severe Errors.
************* EnergyPlus Sizing Error Summary. During Sizing: 2 Warning; 0 Severe Errors.
************* EnergyPlus Completed Successfully-- 7 Warning; 0 Severe Errors; Elapsed Time=00hr 07min 35.94sec…
A repository of generic or complex examples.
Example 01: Attractor Values
ND_001_AttractorValues.gh
Example 02: Curve Values
ND_002_CurveValues.gh
Example 03: Point Attractor
ND_003_PointAttract
h tubes are redundant so surfaces overlap instead of interpenetrate, so it is not a good system.
Cocoon is the best answer these days unless you can get Exowire/Exoskelton to work. If you want more control over shape, feed your uncapped tubes into Cocoon as meta-surfaces and delete any and all of the inner meshes to just keep the outer single closed one, but this is just duplicate-culled lines used as meta-lines:
Turn down the CS input to 0.005 for this result, from 0.02 used for faster preview. In fact bake the lines and only test Cocoon on a few of them in order to get the result you want before doing the whole thing.
Whole thing at 0.005 cell size takes 5 minutes for Cocoon and 2 minutes for refinement to a smooth and even mesh.
Actually, seems like 0.005 is way too fine, giving a 600MB STL file.
So, 0.01 cell size at less than a minute total:
159MB STL which is still a bit too big for places like Shapeways. Wow. OK then 0.02 cell size, but I have to increase diameter or my two smoothing steps in refine collapse things too much, an in fact I set it to no smoothing, getting more volume and a reasonable 46MB STL file:
Alas, now it's more frail and overly organic rather than mechanical. Increasing diameter just merges it into perforated plates too much. File size is simply an issue with this complexity level, so different 3D printing services will have different file size limits.
Exowire/Exoskeleton would work but your original mesh hasn't been MeshMachine remeshed to be regular, so short segments ruin it. Here is just a corner:
I think that's why more wires fails, at least. Pretty temperamental component.
Switching to MeshMachine is needed, I guess, instead of Cocoon refine, to remesh away so many small triangles along the boring tubes. Crucial for good remeshing was to set Flip to 0 or I failed to get a rough enough mesh.
It's an adaptive mesh so I can retain good detail while roughing out the tubes.
MeshMachine is terribly slow for this whole thing, like 6 minutes, and blows up for this overly rough setting, 20 steps, so less rough, ugh, I'm out of time. I think free Autocad Meshmixer is the way to make a better smaller mesh, after a refined output from Cocoon. MeshMachine is just too slow to tweak and when it blows up, creating massive triangles jutting out, it hangs too when you change settings.
Starting with a Cocoon refined mesh certainly helped Meshmixer. Using triangle budget lets me have full control. Here is 150K triangles instead of 200K:
STL file size down to 40MB. I think Shapeways is 70 or 100MB limit? So it can be even finer. Here is the Cocoon output versus the Meshmixer reduction:
To use Meshmixer, turn on View > Show Wireframe, Command-S to select all and use Edit > Reduce from the palette that appears.
Cocoon can end up making a few inner meshes where things get weird in your uneven original mesh with small holes so fish out the main mesh by adding a List Item node.
The best strategy for Cocoon is indeed to make an overly fine STL so you avoid any need to tweak forever in Grasshopper, but then you can achieve a smaller mesh file size while preserving shape instead of things turning all smearly organic in Grasshopper.…
teraction for its Correlations cycle, AA Athens Visiting School scales up its design intentions in order to investigate links among discrete individual architectural systems in its 2013 version, Recharged.
Recharged with interconnectivity on different levels, the theme of investigation will revolve around the design of semi-independent design prototypes acting together to form elaborate unified results. The driving force in Cipher City: Recharged is the synergistic effect behind complex form-making systems where interactive design patterns arise out of a multiplicity of relatively simple rules.
In collaboration with the National Technical University of Athens, Cipher City: Recharged will explore participatory design and active engagement modeling and will continue building novel prototypes upon horizontal planes.
As in 2012, the design agendas of AA Athens and AA Istanbul Visiting Schools will directly create feedback on one another, allowing participation in either one or both Programmes.
Discounts
The AA offers several discount options for participants wishing to apply as a group or participants wishing to apply for both AA Istanbul and AA Athens Visiting Schools:
1. Standard application
The AA Visiting School requires a fee of £695 per participant, which includes a £60 Visiting Membership. If you are already a member, the total fee will be reduced automatically by £60 by the online payment system. Fees are non refundable.
2. Group registration
For group applications, there will be a range of discounts depending on the number of people in the group. The discounted fee will be applied to each individual in the group.
Type A. 3-6 people group: £60 (AA Membership fee) + 635*0.75 = £536.25 (25 %) Type B. 6-15 people group: £60 + 635*0.70 = £504.5 (30%) Type C. more than 15 people group: £60 + 635*0.65 = £472.75 (35%)
3. Participants attending both AA Istanbul and AA Athens | 40% discount
For people wishing to attend both AA Istanbul 2013 and AA Athens 2013, a discount of 40% will be made for each participant. (The participant will pay the £60 membership fee only once.)
£60 (AA Membership fee) + (635*0.60)*2 = £822
For more information in discounts, please visit:
http://ai.aaschool.ac.uk/athens/portfolio/discounts-2013/
Applications
The deadline for applications is 11 March 2013. A portfolio or CV is not required, only the online application form and payment. The online application can be reached from:
http://www.aaschool.ac.uk/STUDY/VISITING/athens…
Added by elif erdine at 12:33pm on December 13, 2012
ng is deciding how and where to store your data. If you're writing textual code using any one of a huge number of programming languages there are a lot of different options, each with its own benefits and drawbacks. Sometimes you just need to store a single data point. At other times you may need a list of exactly one hundred data points. At other times still circumstances may demand a list of a variable number of data points.
In programming jargon, lists and arrays are typically used to store an ordered collection of data points, where each item is directly accessible. Bags and hash sets are examples of unordered data storage. These storage mechanisms do not have a concept of which data comes first and which next, but they are much better at searching the data set for specific values. Stacks and queues are ordered data structures where only the youngest or oldest data points are accessible respectively. These are popular structures for code designed to create and execute schedules. Linked lists are chains of consecutive data points, where each point knows only about its direct neighbours. As a result, it's a lot of work to find the one-millionth point in a linked list, but it's incredibly efficient to insert or remove points from the middle of the chain. Dictionaries store data in the form of key-value pairs, allowing one to index complicated data points using simple lookup codes.
The above is a just a small sampling of popular data storage mechanisms, there are many, many others. From multidimensional arrays to SQL databases. From readonly collections to concurrent k-dTrees. It takes a fair amount of knowledge and practice to be able to navigate this bewildering sea of options and pick the best suited storage mechanism for any particular problem. We did not wish to confront our users with this plethora of programmatic principles, and instead decided to offer only a single data storage mechanism.*
Data storage in Grasshopper
In order to see what mechanism would be optimal for Grasshopper, it is necessary to first list the different possible ways in which components may wish to access and store data, and also how families of data points flow through a Grasshopper network, often acquiring more complexity over time.
A lot of components operate on individual values and also output individual values as results. This is the simplest category, let's call it 1:1 (pronounced as "one to one", indicating a mapping from single inputs to single outputs). Two examples of 1:1 components are Subtraction and Construct Point. Subtraction takes two arguments on the left (A and B), and outputs the difference (A-B) to the right. Even when the component is called upon to calculate the difference between two collections of 12 million values each, at any one time it only cares about three values; A, B and the difference between the two. Similarly, Construct Point takes three separate numbers as input arguments and combines them to form a single xyz point.
Another common category of components create lists of data from single input values. We'll refer to these components as 1:N. Range and Divide Curve are oft used examples in this category. Range takes a single numeric domain and a single integer, but it outputs a list of numbers that divide the domain into the specified number of steps. Similarly, Divide Curve requires a single curve and a division count, but it outputs several lists of data, where the length of each list is a function of the division count.
The opposite behaviour also occurs. Common N:1 components are Polyline and Loft, both of which consume a list of points and curves respectively, yet output only a single curve or surface.
Lastly (in the list category), N:N components are also available. A fair number of components operate on lists of data and also output lists of data. Sort and Reverse List are examples of N:N components you will almost certainly encounter when using Grasshopper. It is true that N:N components mostly fall into the data management category, in the sense that they are mostly employed to change the way data is stored, rather than to create entirely new data, but they are common and important nonetheless.
A rare few components are even more complex than 1:N, N:1, or N:N, in that they are not content to operate on or output single lists of data points. The Divide Surface and Square Grid components want to output not just lists of points, but several lists of points, each of which represents a single row or column in a grid. We can refer to these components as 1:N' or N':1 or N:N' or ... depending on how the inputs and outputs are defined.
The above listing of data mapping categories encapsulate all components that ship with Grasshopper, though they do not necessarily minister to all imaginable mappings. However in the spirit of getting on with the software it was decided that a data structure that could handle individual values, lists of values, and lists of lists of values would solve at least 99% of the then existing problems and was thus considered to be a 'good thing'.
Data storage as the outcome of a process
If the problems of 1:N' mappings only occurred in those few components to do with grids, it would probably not warrant support for lists-of-lists in the core data structure. However, 1:N' or N:N' mappings can be the result of the concatenation of two or more 1:N components. Consider the following case: A collection of three polysurfaces (a box, a capped cylinder, and a triangular prism) is imported from Rhino into Grasshopper. The shapes are all exploded into their separate faces, resulting in 6 faces for the box, 3 for the cylinder, and 5 for the prism. Across each face, a collection of isocurves is drawn, resembling a hatching. Ultimately, each isocurve is divided into equally spaced points.
This is not an unreasonably elaborate case, but it already shows how shockingly quickly layers of complexity are introduced into the data as it flows from the left to the right side of the network.
It's no good ending up with a single huge list containing all the points. The data structure we use must be detailed enough to allow us to select from it any logical subset. This means that the ultimate data structure must contain a record of all the mappings that were applied from start to finish. It must be possible to select all the points that are associated with the second polysurface, but not the first or third. It must also be possible to select all points that are associated with the first face of each polysurface, but not any subsequent faces. Or a selection which includes only the fourth point of each division and no others.
The only way such selection sets can be defined, is if the data structure contains a record of the "history" of each data point. I.e. for every point we must be able to figure out which original shape it came from (the cube, the cylinder or the prism), which of the exploded faces it is associated with, which isocurve on that face was involved and the index of the point within the curve division family.
A flexible mechanism for variable history records.
The storage constraints mentioned so far (to wit, the requirement of storing individual values, lists of values, and lists of lists of values), combined with the relational constraints (to wit, the ability to measure the relatedness of various lists within the entire collection) lead us to Data Trees. The data structure we chose is certainly not the only imaginable solution to this problem, and due to its terse notation can appear fairly obtuse to the untrained eye. However since data trees only employ non-negative integers to identify both lists and items within lists, the structure is very amenable to simple arithmetic operations, which makes the structure very pliable from an algorithmic point of view.
A data tree is an ordered collection of lists. Each list is associated with a path, which serves as the identifier of that list. This means that two lists in the same tree cannot have the same path. A path is a collection of one or more non-negative integers. Path notation employs curly brackets and semi-colons as separators. The simplest path contains only the number zero and is written as: {0}. More complicated paths containing more elements are written as: {2;4;6}. Just as a path identifies a list within the tree, an index identifies a data point within a list. An index is always a single, non-negative integer. Indices are written inside square brackets and appended to path notation, in order to fully identify a single piece of data within an entire data tree: {2,4,6}[10].
Since both path elements and indices are zero-based (we start counting at zero, not one), there is a slight disconnect between the ordinality and the cardinality of numbers within data trees. The first element equals index 0, the second element can be found at index 1, the third element maps to index 2, and so on and so forth. This means that the "Eleventh point of the seventh isocurve of the fifth face of the third polysurface" will be written as {2;4;6}[10]. The first path element corresponds with the oldest mapping that occurred within the file, and each subsequent element represents a more recent operation. In this sense the path elements can be likened to taxonomic identifiers. The species {Animalia;Mammalia;Hominidea;Homo} and {Animalia;Mammalia;Hominidea;Pan} are more closely related to each other than to {Animalia;Mammalia; Cervidea;Rangifer}** because they share more codes at the start of their classification. Similarly, the paths {2;4;4} and {2;4;6} are more closely related to each other than they are to {2;3;5}.
The messy reality of data trees.
Although you may agree with me that in theory the data tree approach is solid, you may still get frustrated at the rate at which data trees grow more complex. Often Grasshopper will choose to add additional elements to the paths in a tree where none in fact is needed, resulting in paths that all share a lot of zeroes in certain places. For example a data tree might contain the paths:
{0;0;0;0;0}
{0;0;0;0;1}
{0;0;0;0;2}
{0;0;0;0;3}
{0;0;1;0;0}
{0;0;1;0;1}
{0;0;1;0;2}
{0;0;1;0;3}
instead of the far more economical:
{0;0}
{0;1}
{0;2}
{0;3}
{1;0}
{1;1}
{1;2}
{1;3}
The reason all these zeroes are added is because we value consistency over economics. It doesn't matter whether a component actually outputs more than one list, if the component belongs to the 1:N, 1:N', or N:N' groups, it will always add an extra integer to all the paths, because some day in the future, when the inputs change, it may need that extra integer to keep its lists untangled. We feel it's bad behaviour for the topology of a data tree to be subject to the topical values in that tree. Any component which relies on a specific topology will no longer work when that topology changes, and that should happen as seldom as possible.
Conclusion
Although data trees can be difficult to work with and probably cause more confusion than any other part of Grasshopper, they seem to work well in the majority of cases and we haven't been able to come up with a better solution. That's not to say we never will, but data trees are here to stay for the foreseeable future.
* This is not something we hit on immediately. The very first versions of Grasshopper only allowed for the storage of a single data point per parameter, making operations like [Loft] or [Divide Curve] impossible. Later versions allowed for a single list per parameter, which was still insufficient for all but the most simple algorithms.
** I'm skipping a lot of taxonometric classifications here to keep it simple.…
Added by David Rutten at 2:22pm on January 20, 2015
lly it should not make much of a difference - random number generation is not affected, mutation also is not. crossover is a bit more tricky, I use Simulated Binary Crossover (SBX-20) which was introduced already in 1194:
Deb K., Agrawal R. B.: Simulated Binary Crossover for Continuous Search Space, inIITK/ME/SMD-94027, Convenor, Technical Reports, Indian Institue of Technology, Kanpur, India,November 1994
Abst ract. The success of binary-coded gene t ic algorithms (GA s) inproblems having discrete sear ch sp ace largely depends on the codingused to represent the prob lem variables and on the crossover ope ratorthat propagates buildin g blocks from pare nt strings to childrenst rings . In solving optimization problems having continuous searchspace, binary-co ded GAs discr et ize the search space by using a codingof the problem var iables in binary st rings. However , t he coding of realvaluedvari ables in finit e-length st rings causes a number of difficulties:inability to achieve arbit rary pr ecision in the obtained solution , fixedmapping of problem var iab les, inh eren t Hamming cliff problem associatedwit h binary coding, and processing of Holland 's schemata incont inuous search space. Although a number of real-coded GAs aredevelop ed to solve optimization problems having a cont inuous searchspace, the search powers of these crossover operators are not adequate .In t his paper , t he search power of a crossover operator is defined int erms of the probability of creating an arbitrary child solut ion froma given pair of parent solutions . Motivated by t he success of binarycodedGAs in discret e search space problems , we develop a real-codedcrossover (which we call the simulated binar y crossover , or SBX) operatorwhose search power is similar to that of the single-point crossoverused in binary-coded GAs . Simulation results on a number of realvaluedt est problems of varying difficulty and dimensionality suggestt hat the real-cod ed GAs with t he SBX operator ar e ab le to perform asgood or bet t er than binary-cod ed GAs wit h t he single-po int crossover.SBX is found to be particularly useful in problems having mult ip le optimalsolutions with a narrow global basin an d in prob lems where thelower and upper bo unds of the global optimum are not known a priori.Further , a simulation on a two-var iable blocked function showsthat the real-coded GA with SBX work s as suggested by Goldberg
and in most cases t he performance of real-coded GA with SBX is similarto that of binary GAs with a single-point crossover. Based onth ese encouraging results, this paper suggests a number of extensionsto the present study.
7. ConclusionsIn this paper, a real-coded crossover operator has been develop ed bas ed ont he search characte rist ics of a single-point crossover used in binary -codedGAs. In ord er to define the search power of a crossover operator, a spreadfactor has been introduced as the ratio of the absolute differences of thechildren points to that of the parent points. Thereaft er , the probabilityof creat ing a child point for two given parent points has been derived forthe single-point crossover. Motivat ed by the success of binary-coded GAsin problems wit h discrete sear ch space, a simul ated bin ary crossover (SBX)operator has been develop ed to solve problems having cont inuous searchspace. The SBX operator has search power similar to that of the single-po intcrossover.On a number of t est fun ctions, including De Jong's five te st fun ct ions, ithas been found that real-coded GAs with the SBX operator can overcome anumb er of difficult ies inherent with binary-coded GAs in solving cont inuoussearch space problems-Hamming cliff problem, arbitrary pr ecision problem,and fixed mapped coding problem. In the comparison of real-coded GAs wit ha SBX operator and binary-coded GAs with a single-point crossover ope rat or ,it has been observed that the performance of the former is better than thelatt er on continuous functions and the performance of the former is similarto the lat ter in solving discret e and difficult functions. In comparison withanother real-coded crossover operator (i.e. , BLX-0 .5) suggested elsewhere ,SBX performs better in difficult test functions. It has also been observedthat SBX is particularly useful in problems where the bounds of the optimum
point is not known a priori and wher e there are multi ple optima, of whichone is global.Real-coded GAs wit h t he SBX op erator have also been tried in solvinga two-variab le blocked function (the concept of blocked fun ctions was introducedin [10]). Blocked fun ct ions are difficult for real-coded GAs , becauselocal optimal points block t he progress of search to continue towards t heglobal optimal point . The simulat ion results on t he two-var iable blockedfunction have shown that in most occasions , the sea rch proceeds the way aspr edicted in [10]. Most importantly, it has been observed that the real-codedGAs wit h SBX work similar to that of t he binary-coded GAs wit h single-pointcrossover in overcoming t he barrier of the local peaks and converging to t heglobal bas in. However , it is premature to conclude whether real-coded GAswit h SBX op erator can overcome t he local barriers in higher-dimensionalblocked fun ct ions.These results are encour aging and suggest avenues for further research.Because the SBX ope rat or uses a probability distribut ion for choosing a childpo int , the real-coded GAs wit h SBX are one st ep ahead of the binary-codedGAs in te rms of ach ieving a convergence proof for GAs. With a direct probabilist ic relationship between children and parent points used in t his paper,cues from t he clas sical stochast ic optimization methods can be borrowed toachieve a convergence proof of GAs , or a much closer tie between the classicaloptimization methods and GAs is on t he horizon.
In short, according to the authors my SBX operator using real gene values is as good as older ones specially designed for discrete searches, and better in continuous searches. SBX as far as i know meanwhile is a standard general crossover operator.
But:
- there might be better ones out there i just havent seen yet. please tell me.
- besides tournament selection and mutation, crossover is just one part of the breeding pipeline. also there is the elite management for MOEA which is AT LEAST as important as the breeding itself.
- depending on the problem, there are almost always better specific ways of how to code the mutation and the crossover operators. but octopus is meant to keep it general for the moment - maybe there's a way for an interface to code those things yourself..!?
2) elite size = SPEA-2 archive size, yes. the rate depends on your convergence behaviour i would say. i usually start off with at least half the size of the population, but mostly the same size (as it is hard-coded in the new version, i just realize) is big enough.
4) the non-dominated front is always put into the archive first. if the archive size is exceeded, the least important individual (the significant strategy in SPEA-2) are truncated one by one until the size is reached. if it is smaller, the fittest dominated individuals are put into the elite. the latter happens in the beginning of the run, when the front wasn't discovered well yet.
3) yes it is. this is a custom implementation i figured out myself. however i'm close to have the HypE algorithm working in the new version, which natively has got the possibility to articulate perference relations on sets of solutions.
…