Finally found You !

[Varpu]

Hi everyone !
Since this is my first posting here I think a brief introduction to myself is appropriate:

I am the owner and the administrator of a site called UTCMS. It is a tiny server humming in the corner of one of the rooms in our home.

That server has currently over 9500 different Unreal nad UT maps in it. These are unique maps - no duplicates.

The user interface is crappy but - well - I would not even dream of getting even near the 1337:ness of what You can see here at Unreal Playground. I must admit that my first impression of UP made my jaw drop. Now - if they only had a more accuracy on their search mechanism....

[SK TastesLike KFC]

The site is still evolving and eventually all the bugs will be worked out. You could PM MadMikey and convey your concerns, desires etc. I KNOW he would be glad to hear from you and eager to go over any ideas you may have.

Tell him KFC sent you.


It's great to have you here, Welcome







SKTLKFC:D

[StealthDP]

Welcome Varpu!

Yes, we are working on numerous new features, including a more advanced search mechanism. The system in place over at your site is awesome, you did a great job on it.

Again, welcome to the Playground...

[D.J.P.]

Hey Varpu, UTCMS is a great site. I use it for getting well.....anything especially the texture packs.

[Freakish]

Welcome to the Playground!

Yeah..UTCMS is my backup whenever I need anything. And definitely give us suggestions!

[ShirleyFT]

Hi Varpu(no. 500). Welcome.

[Varpu]

Hmmm... there are several things that are missing from most of the UT sites (including UTCMS):

# Ability to check the package contents, missing files, unsuitable file names, file names that have already been taken and so on BEFORE importing the map into the system. UTCMS does this check during the import but it is so to say "too late". Ofcourse this would involve the author to send the package for inspection and when the files would be in "allright" condition (s)he would give it a "go" or "no-go". Some sort of halfway stop before actually pushing the map package out of the tube.

# Ability to create uz files on the fly. This could be done so that the web server detects it when a certain directory is accessed by a package like http://furpile.com/uzfiles/Citytex.utx.uz (don't klick that link, there is nothing in there). For example this could be very easy to implement at UTCMS as it already has the packages stored in separate components. Any UT server administrator knows what is the value of this kind of a service.

# Ability to create UMOD files on the fly. Even if this is not the most important issue, it can make living with different files a lot easier. It can also make living hard as the UMOD installers replace Your files without checking.

I am in progress of making an automatic map classifier that uses neural-network and creates a keyword list for each component. It will inspect the material that were used to create the map. This in turn can be used for more intelligent searches over the material. I can create a dedicated access to this service once it is up and running.

I can also give/create access to other dedicated services. One of this kind would be a non-HTTP search engine that creates text output of the given search criteria. Simplest form would be polling existence of a certain thing. The HTTP interface used at UTCMS is not too suitable for mirroring etc....

Generally speaking - I would like to see more cooperation between sites. For example now the mapper has to upload his/hers stuff to several sites to gain coverage - this all including the updates (Yeah, at UTCMS You can update Your maps). Think about it: Put my map to one site and it WILL spread itself to all the other sites as well. If You want this kind of functionality I can write the spesification for it - something I do daily to earn the butter over my bread.

If You need anything - You know what to do (esko.ilola@sonera.com)

[ShirleyFT]

That map classifier sounds cool. I'm still a student. I was wondering about the neural network. How are you setting up the input layer? Does it use individual inputs for maps, textures, system, etc... or are you breaking the input down to smaller groups, types of maps, types of textures? If you're using the latter, do you use any open inputs that may be filled in later once new materials are found?

It sounds interesting.

[Varpu]

It will use all available information. As You propably know, one of the key features of this animal is that it becomes more accurate as the input data variety grows. The problem is ofcourse to make the input data into "digital" form.

I have already implemented most of this thing. It uses self organizing map because giving exact input values would be too hard. As a matter in fact, the NN gives out the most likely and the immediate neighbors of it. The first 100 maps will be hard as I need to "teach" the thing to regognize patterns. After this it will become more and more accurate each time. The NN data (the matrix) is persistent.

[ShirleyFT]

Thanks, Varpu. You've already answered my second question which was about the training of the NN. I guess with a library of 9500 maps you can easily pick 100 which would represent a wide variety. I think I know what you're doing now and thanks.

[Varpu]

LOL - I just ran a few tests with this NN thing. It classified the CTF-Face with "urban", "castle", "sniping" and "space" keywords ... not too bad...
Now, if I only could figure out how to make it generate reviews something like:

CTF-Face:

This urban map which is located in space suits well for sniping. The medieval themed castle.....

The funny side in here is that I REALLY don't know how the thing came into conclusion of giving those keywords.

[Chrysaor]

Programming is so cool. I'm glad I horrible at it.

[Freakish]

Hmm.. I'm gonna put some AI courses on the agenda..

Cool stuff Varpu.

[ShirleyFT]

Yes, Freak. You're there at Berkeley, load up with the AI. You should also check out the machine vision courses.

My friend Jesse will be there next semester. You can't miss him. He always wears a psychedelic t-shirt with a big green frog on it. He's a physics/comp sci major and will be doing graduate work in quantum computing.

[Freakish]

Originally posted by ShirleyFT
Yes, Freak. You're there at Berkeley, load up with the AI. You should also check out the machine vision courses.

My friend Jesse will be there next semester. You can't miss him. He always wears a psychedelic t-shirt with a big green frog on it. He's a physics/comp sci major and will be doing graduate work in quantum computing.

uh..you don't think he'll stand out do you?:eek:

[HortonsWho]

VARPU!!!

Welocme! Glad you are here!

Hort