ikinci bir saldırı | another forage into evo-computation

a1. NURBS yüzeyleri	a2. tesadüfi ilk kuşak	a3. temel bileşenler, 5 eğri: düz, basık, orta, yüksek, dik
b1. bir “aile benzerliği”ne karşılık çeşitlilik / gürültü	b2.	b3.
c1.	c2.	c3.

[sanıyorum ekim sonlarına kadar bununla uğraşmaktaydım (uğraşılmaktaydı).... işte rapor:]

ikinci deney serisi: maya-mel ortamında 3 boyutlu deney: lokal kontrol, gürültü, sözel tanımlar ::

ikinci bir deney serisi ise daha gelişmiş bir uygulama üzerinden üç boyutlu yüzeylere yönelik olarak kurgulandı. bu deneyin de esas amacı scripting, 3 boyutlu geometri ve pratikte evrimsel hesaplamalara yönelik bilgi ve beceriyi geliştirmekti [bu da doğru, yannız, bu scripte baya zaman harcadım, boşa gitti o zaman, bikaçbin satır kod yazıp MEL ile ölesiye boğuştuktan sonra kodun çoğunun tekrar kullanılamayacağını farkettim.. ondan sonra dedim ki ya araştırmacı, git adam gibi bir dil kullan, doğru bütün bir program yaz, en azından temel bileşenleri sonraki her deneyinde kullanırsın, işe yarar veri yapıların olur, hazır kütüphaneleri elinin altına indirirsin...]. bu deney serisinde genotip temsil yaklaşımı geliştirilerek, “haritalama” ve “arama” aşamaları ayrıldı ve “çaprazlama” (crossover) gibi opsiyonlar da uygulamaya katıldı, ayrıca bir serinin sonuçlarının diğer seri için kolayca girdi oluşturabilmesi sağlandı.
deneyin ana teması önceden tanımlanmış bir seri eğri ile kontrol edilen yüzeylerde çeşitli yüzey özelliklerini oluştururken çeşitlilik ve gürültü düzeyini kaybetmemekti [bir baktım da janssen, bu projeye başlarken temel ilham kaynağım olan janssen, bu "aile benzerliğine rağmen çeşitlilik" hususunu yazmış satır aralarında]. oluşturulan bütün yüzeyler beş çeşit eğriden oluşuyordu. her bir eğrinin hem bir ismi hem de bir numarası bulunuyordu. İsimler kontrol mekanizmasını basitçe tanımlanabilir ve anlaşılır hale getirirken (1 : düz, 5 : dik, 2 : basık vd..) numaralar genotipin temel bileşenlerini (genler) oluşturuyordu. sonuç olarak ‘arama’ genotipte oluşan örüntülerin incelenmesi, bazı örüntülerin diğerlerine nazaran tercih edilmesi şeklinde gerçekleşti. sözgelimi bir ‘kromozom’da oluşan : { *, *, 1, 3, 1, 3, 1 } örüntüsü (bu örnekte ilk iki gen belirleyici değil) dalgalı bir yapıya karşılık geliyor. basit bir ’dalgalı’ yüzey araması {*, *, x, x+-a, x, x+-a, x} türünden bir sorgulamayla gerçekleştirilebilir. uygulamada daha ‘gürültülü’ aramalar tanımlandı. ‘artan’, ‘dalgalı’, ‘basık’ vd yüzey aramaları ve bunların çeşitli seri bileşimlerinden oluşan bileşik aramalar gerçekleştirildi.
örüntü aramasının ve bu aramaların insani dilde basitçe tarif edilebilir olmasının (semantik karşılıklar) genel olarak bütün araştırma boyunca önemli bir yer tutacağı görülüyor. arama/değerlendirme yaklaşımı ve üretim yaklaşımı bir ‘itme-çekme düzeneği’ oluşturuyor [araştırmacı mısın mimar mısın ba! itmeçekmeymiş...]. değerlendirme ve puanlama bir seçilim baskısı yaratarak ‘yönlendirme’ etkisi uygularken fenotipin ve arama uzayının tanımlanma ve mutasyonlar üzerinden dönüştürülme tarzı popülasyonun o yöne doğru dönüşebilmesini sağlıyor. İki yaklaşımın bir arada etkili kullanımı arama uzayının istenen yönde daraltılmasını ve bu daralmaya rağmen yeterince geniş, beklenmedik ürünler verebilecek bir durumda tutulabilmesini sağlayabilir. [raporlarda bi şekilde laubali olmayan bir dille anlatmak gerekiyordu, o kadar anlatmışken de burda silbaştan yazamadım... affola..]

ilk evrimsel sikriptlerim bunla meşguldü | my first evo-scripts were occupied with these

a1. tipik bir ilk kuşak (10 noktadan oluşan NURBS çizgileri)	a2. ortogonal, son jenerasyon	a3. ortogonal
b1. ortogonal (yerel optimaya takılmış)	b2. ortogonal	b3. ortogonal
c1. poligonal (500. jenerasyon)	c2. poligonal (1000. jenerasyon)	c3. ortogonal + poligonal (hibrit)

imajlar iyi görünmüyorsa (ki görünmüyor) bu linki takip...

[bunlar da raporuma (birinci tekil şahıs hiç varolmamışçasına) yazdıklarım:]

ilk deney serisi: maya-mel ortamında 2 boyutlu deney: lokal, açısal kontrol-kontrolsüzlük

ilk “deney” olarak maya-mel scripting ortamında basit bir evrimsel algoritma kodlandı. başlangıçta rastgele oluşturulan bir seri lineer iki boyutlu düzenlemenin her kontrol noktasında oluşan lokal açısal ilişkiler [o noktada kesişen vektörlerin arasındaki açıları ölçtüm demek istiyorum] incelenerek açıların toplamı istenen kriterlerle karşılaştırıldı [kriter dediğim de 90'ın katları mı gibi basit bir soru]. deneyin esas amacı basit bir evrimsel algoritma oluşturarak scripting becerilerini geliştirmek ve evrimsel hesaplamalar uygulamalarına giriş yapmaktı [doğru bu], ancak sonuçlar global ve lokal değerlendirmeler konusunda bir takım fikirler de sağlıyor [her musibetten...].

değerlendirme kriterleri kuvvetli, sade ve tek yönlü olunca sistemin popülasyonu istenen açısal ilişkileri taşıyan bireylere doğru kolayca “evrilttiği” gözleniyor (b3’e kadar örnekler). ancak sadece açıların toplamına yönelik puanlama yapıldığı için görece düzenli ve basit evrimler gerçekleşiyor ve sistem önceden kolayca öngörülemeyen “eğilimler” veya “tercihler” gösteriyor. dolayısıyla böyle bir sistem belirli hedeflere yönelik olarak kullanılacaksa daha güçlü kontrol sağlayan yaklaşımların kullanılması gerekiyor.

takip eden deneyde hem açıların toplamına hem de global bazı özelliklere yönelik puanlamalar yapıldı [aslında hesaplarda hata yapmışım :]], ortaya çıkan sonuçların ilk bakışta çok daha kaotik bir görünüm arzettiği söylenebilir (c1, c2) [neden acaba..], ancak esasında çoğu “birey” aranan özellikleri bir ölçüde taşıyor [yanlışlıkla aranan...]. fakat daha “anlamlı” ve kolay tanınır biçimsel düzenlemelere ulaşmak için sadece “lokal” düzeyde işleyen değerlendirmelerin yeterli olmadığı, “global” biçime yönelik daha güçlü seçim kriterleri uygulamak gerekeceği anlaşılıyor.

üçüncü olarak bir tür evrimin diğer bir tür evrim için girdi oluşturduğu iki aşamalı sistemler ise beklenmedik, anlaşılabilir ve düzgün sonuçlar verebiliyor (c3). bunun sebebi ilk iki deneyden farklı olarak ikinci aşamada evrime daha iyi tanımlanmış, daha homojen bir ilk jenerasyonla (başlangıç kütüphanesi) başlanıyor olması olabilir.

ilk deney serisinin sonuçları evrimsel algoritmalarla istenen bir biçimsel düzenleme evriltebilmek/üretebilmek için, bir ya da birden çok aşamada ama her durumda farklı ‘düzey’lerde, hiyerarşik olarak işleyen kriter kümeleri ve değerlendirme yaklaşımları kullanılmasının etkili olabileceğini düşündürüyor. [hmm]

"yapmasa mıydık, etmese miydik" işte sen busun!

geçen yıl için hazırlayalım dediğimiz birinci sınıf stüdyosu yayınıyla ilgili düşünüyorum, bir yazı yazarsam yaptıklarımız değil de, yapmadıklarımız üzerine yazsam diye (yapsa mıydık yani?) ... bir de bol bol çocukların işlerini koymak lazım, altına da uzun uzun orda ne yaptıklarını yazmak lazım...

at that | orda hak verdim çocuğa

i (have?) attended a, so called (appropriate?), colloquyum, collokiyum, kollokyum, qolloquium? öeeh wadevır, it was nice, so nice that it was worth spending all afternoon on it, in it, with it, around it, wadevır again... the event was organized to (on, towards, at?) the honour of another more important event (that was bige's phd defence) by, let's say the unit, design informatics, which i am currently enrolled in, as a guest researcher. here this unit is under the title of building technology! if it were in my home-institution, this wouldn't be the case, but this is a reason why i wanted to come here, and if somehow i were (was?) to work on (in, at, around..) academics, why i would prefer to go back (oh what a complex relationship with an academic institution (home-inst.)one of those love-hate smtg (will you dare?)...)

i wasn't (haven't been) thinking about those reasons for the time being (fail again! no! not the best usage man(dare?)).. this unit is working mainly on performance based design, or an integral approach to design where "rational" performances of a to be(just write "prospective" as an adventure!) built artifact is meant to be used as an important driver of the design process. here, performance means, environmental, ecologial and sometimes structural performance (this is a rough generalization of course).

OK. what's performance? which performance? what about a minimum design task, to be counted as architectural design? describe me a minimum design task and minimum requirements to be answered, if the solution is to be counted as a solution. oh you won't. i will find out. or better: i will narrate.

now i want to combine vagaries of a designer and current rationalities of economics, environmental comfort or ecological sensitivities (sensibilities you fool!) and make (bake?) out of this dough (come and utter this word!) a nice spongy(is this the word you've found as a substitute to "gevrek"? the opposite? go back to your country to where you belong, gow-back joh-jo) bread. just as a human designer does. it will, no doubt, be a caricature. but that's okay. we all like humor. (but not a cynical humor!)

i was talking to a friend (jurrian), i told him that, watching and watching all those performances, all the presentations, conferences, collo(smtg)s, lectures etc. they only become meaningful when you are also producing something yourself, and he said, you have to balance those things (watching and doing), he had a point in that (on that, at that) at that.

_prof. sariyildiz described her unit's recent phd projects in contrast with projects just focussing on fancy forms. the message was: there are serious challenges, facing humanity, and architects (like environmental exicengies), and formal issues are not mere free-form excercises. she was deliberately omitting a good portion of the story (i.e. marketing aspects) but this omission is the basic tenet of her stance. the unit (which i am visiting) is re-describing form generation in terms of rational performances, in an integral manner (sounds quite like form-follows-function eh?), hence a nice paper-opponent for me to better forge my stance (which has a lot to do with past years' anti-professionalist studio experiences and also patrick janssen's (who's in part following lawson i guess) seemingly simple but for me important theoretical underpinnings) out of... (long sentence) (saying out loud the obvious) (so now let's shout:) any architectural design process is formed out of both, in a way, rationalized or substantiated acts, and also vagaries (or for janssen, "preconceptions"), none to be missing. form has never been totally a function of function. but, you know, it was because the means were not developed yet. until now we weren't able to assess an important portion of the performances of a building beforehand. perhaps, from now on, form might become a function of several performances? it's easy to believe it in netherlands, cause they generally tend to keep it really simple. but wait a minute, we seem to be omitting something here!?

_prof. krishnamurti was here, and presented their latest work on parametric BIM(?). at least for simple buildings (with definable floors and cornered plan layouts), using revit, they produced a system that measures the building performance in terms of BREAM, or LEED. nice tool, nice work. he was suspicious of the adequacy of LEED but it is the current standard for many countries. he also mentioned "design patterns for parametric modelling". so the patterns of alexander came back through the backdoor, only after passing through the realm of programming, where the idea was adopted rather effusively (but here my colleague juse is also using the idea of pattern language, in conjunction with shape grammars for urban design). the most peculiar moment of krishnamurti's presentation was, that, he was also suspicious of all those tools that has continuously been developed tirelessly by the phd researchers, to be used "in the early phases of conceptual design", or to give designers decision support on formal matters, or supply designers with alternative designs so that they can choose, or do the design on behalf of them.. he was quite right at that, designers want to do the enjoying part themselves and would be happy if something manages to do the boring, uncreative jobs. (that's why i'm producing a similar application but never calling it a tool :] i'm well avare that, only i, will be fascinated with using it. and i even don't plan to use it for education, for i could better help the students if i could assist them develop their own ideas with programming skills. what's the justification of forcing a studioful-of-students to use your own program [or idea] for designing?) actually that remark was in complete opposition to the work of some academics that were present :] but the man has a nice and funny air, and he's so venerable and established a researcher in the field, that it didn't bring a cold breeze at berlagezaal at all (which was already cold by the way.)

_joop paul, a professor at the university and an engineer of ARUP at the same time (it's not surprising that it's possible here), showed slides of a hi-rise. which had a slender prefabricated (as i understood) structure which didn't cost more than conventional methods and forms of tower building. the message is: by using computational tools effectively, you can stay both feasible and creative. another point on behalf of performance based design? but a hi-rise has always been an application where structural performance (and circulation and evacuation performance) has a sheer precedence; so this doesn't count? or does it? look at CCTV and similar novel hi-rise projects...

_prof de vries from eindhoven mentioned bayesian belief networks and design decision support systems. actually these tools, as krişnamurti asserted are not for human designers, they are for automated systems, at least i understand it this way, surprisingly, during this week i was searching in their archive of phd dissertations, and i have downloaded several of them.. bayesian belief networks in design decision support eh? niiice!

_and finally, the question of adaptive systems. the "quest" for adaptive systems? (something more than showing the realtime "weather forecast" which is currently surprisingly popular in architectural surfaces!?) adaptive buildings or systems can only be possible by integrated data-form environments. so that form follows data follows form follows... "integral"? "integrated"? when you say integral, it's a puzzling term.

"design games" çalışma hayatına böyle başladı (yavrucak) | thus started the toil of the "design games"

these are selected from the first tests, the only tool is color area tool, and is used randomly, as you can see, but what is a tool?



a tool is one of the four atomic elements of design games system (or also framework). another one is the character. tools do, charcters direct. but in fact the separation is far than clear cut. all the atoms, ultimately, are meant to be interchangeable. so that, from the very beginning, a clear cut separation is neither possible, nor desired. and there are also objects, or individuals, in evolutionary terms. they are the alternative products, or designs that are worked on.

and? the games. a game is a compound [come on what kind of atom is it!?] but it could also be interchangeable with the others [possibly self referential?], let's calm down, right now all atoms are discrete and they are not implemented as interchangeable. a game is formed when a character directs a tool and an action is performed on an object.

and these things are produced, the following are also almost random products of the first runs to obtain a desktop icon for design games itself, but i sincerely don't remember which ones were last generation (i.e: evolved, designed) and which are byproducts of the process:



after a thousand runs i started to identify the defects of the system (it's a continuous refining and re-doing process) and then i conducted another series of runs with better character definitions and larger resolution, following are the designs of design games for itself:




the bright colors are not surprising, our character loved them, and it wanted several color areas and it got them. utilized tools were, pattern stamp, gradient, color area and text (i.e. all of them) so we moved on to another task, producing (or designing if you like) a desktop wallpaper for my laptop. again we started with very low resolution trials, these are collected amongst the byproducts, meaning that the poor evaluation capabilities of the systems or coded likings of the characters (by me) prevented their identification as 'fine':



if you didn't notice the images that were produced on previous runs were fed as patterns into these new runs, and only tool was the pattern stamp. and following are the designed ones, look at the differences between these two sets (you can only see it if you go to their pages):



so we went on and on for nights, i was sleeping and design games was working, as appropriate. check the following, they are collected from the garbage of the final runs for a high-resolution desktop wallpaper, different high resolution patterns are added to the system, and that's all:




end the evolved ones: (i really worked hard to find something fine amongst the evolved ones, there were faults in some analysis mechanisms and character definitions were apparently awful, but these are fine, i like them:)




OK, enough for now. there are other stuff to come. but they will wait for a bit. i have to have some break, and do other stuff, like blogging.

here comes the "design games" | here i am

ok, now we have another paper, no we don't have that paper, we just submitted a paper for another conference. but at least i'm glad that the researcher came to a point to start expressing its project and how it is related with all the blabla about design, games, and all the other things that the researcher had been writing and writing around this bilog. so let's start with a short explanation, i'm abstracting from my paper:

>>“design games” is a framework for a semi-automated design system which uses an evolutionary computation approach as its 'engine'. the ultimate aim of this study is developing an automated system which could take part in several design 'sub-tasks', in a collaborative environment, serving as a team of customizable virtual designers. a computer program is being developed to test and improve the approach. the system that is developed is based on a flexible, open scheme that enables us to tackle sub-tasks pertaining to different design areas, which are chosen amongst the rather ill-defined and wickedly arbitrary ones.
the system is being tested on several graphic design and illustration tasks. currently design games do not have tools developed enough for conducting a full-fledged design process solely on itself, but it manages to cooperate with the researcher and develops itself as a pupil.

[in this entry i will not be writing about the conceptual framework, 'cause they are written here, and i wrote an entry about that here.. and.. i would be glad if you have a look at that, 'cause together with what i will describe in this and next entries, it's quite more meaningful now... and... i will not explain what evolutionary computation is, but you can find something short here]

>>versatility of evolutionary computing:
_evolutionary computation approaches have already been tested in a diverse array of problem solving areas and this diverse array, no doubt, includes several design and art fields as well. this apparent versatility of evolutionary computation as an approach, and its adaptability into very different types of problem solving areas and artistic fields, is one of the reasons why we [the researcher :] utilized this approach in our study. we conceive each design process as fragmented, including a diverse array of different types of actions. hence, a corresponding system tackling a similar design process should be versatile and should be able to run a diverse array of tools pertaining to each different type of sub-tasks. we envisage evolutionary computation practically as the central engine for our system with its basic alternatives-producing (generate) and selection (test) structure.

>>form generation in evolutionary systems
_form generation is one 'vagabond' realm. in a way all design is form generation[!?] but, relationships between aims, reasonings, functionalities and forms are not always well-known; at least when it comes to creative design areas like, graphic design, architecture, interior design, fashion design or product design etc.
_it should be so formidable an area or perhaps, on the contrary so uninteresting a research field that, when it comes to form generation, most of the times, researchers either tended to convert form generation process into a better defined search (by simplifiying it of course), or, following the precedent of "the blind watchmaker" [of richard dawkins] they preferred implementing interactive evolutionary systems, where the alternatives are evaluated by human evaluators. however, interactive evaluation process is time consuming and tiresome (and in a way, it is putting the burden back again on humans).

>> “other performances” and an evolutionary proposal
_"in this study, we chose to start with tackling those irremediably 'wicked-problems', and without trying to reduce or define them in terms of explicit criteria or corresponding rational performances. the phrase “other performances” denotes those aspects or criteria, which we usually cannot 'account for' in explicit or rational terms, but we utilize heavily when we're designing. these are our heuristics, likings, beliefs, prejudices, irrationalities, principles, vagaries, and simply everything that comes handy at the moment. what we (un)define as other performances are generally not researched extensively or seen as resistant to definition. (however, that we started with them doesn't mean that we thought they were enough to define any complicated design process. our supposition is, that, in any complicated design process, different types of design acts are incorporated, and this pertains to different levels of 'definedness' as well.)"

_in order to embody this conception of the other performances, we are substituting "virtual designer characters" in place of human decision makers. they collect and imply humane vagaries heuristics, tastes etc.
_another important factor is 'restricted randomness'. the idea here is, if you don't know the reasoning behind a principle, and if it seems to yield somehow random results then it might have a genuinely random portion and the way to 'imitate' it might be randomness.
_in spite of all the emphasis on noise and randomness, design games is essentially a knowledge intensive system. knowledge is incorporated by means of tools, characters and patterns and if design games will ever be a real designing system it should learn much more than what it currently knows.

[what design games currently doing will be on the next post, for the ones who are really interested, just have a look at these works: ]

bentley, p.j. and corne, d.w. (eds): creative evolutionary systems. academic press, london. (2002)
bentley, p.j. (ed): evolutionary design by computers. morgan kaufmann publishers, san francisco (1999)
caldas, l.g.: generation of energy-efficient architecture solutions applying gene_arch: an evolution-based generative design system. advanced engineering informatics archive volume 22 , issue 1 (2008)
bitterman, m.: intelligent design objects. phd dissertation, technical university delft (2009)
hong, l., mingxi, t.: evolutionary design in a multi-agent design environment. applied soft computing 6 pp207–220 (2006)
janssen, p.h.t.: a design method and computational architecture for generating and evolving building designs. phd dissertation, hong kong polytechnic university (2004)
xiyu, l., mingxi, t., frazer, j.: an eco-conscious housing design model based on co-evolution. advances in engineering software 36 pp115–125 (2005)
banerjee, a., quiroz, j., louis, s.: a model of creative design using collaborative interactive genetic algorithms. in: gero j., goel, a. (eds.): design computing and cognition '08, pp397-416 springer science + business media b.v. (2008)
nishino, h., sueyoshi, t., kagawa,t., utsumiya, k.: an interactive 3d graphics modeler based on simulated human immune system. journal of multimedia, vol. 3, no. 3, july 2008 (2008)
marin, p., bignon, j., lequay, h.: integral evolutionary design, integrated to early stage of architectural design process generative exploration of architectural envelope responding to solar passive qualities