TMD MWS Server

[AmbivalentDuck]

I've been thinking about setting up a TMD MWS server.  Does anyone know how much bandwidth that would take?  It seems like MWS tracks IPs and decklists via the server, so it's almost ideal for setting up some sort of real data analysis on the format.  Of course, any data I gleaned would be completely public and people could opt out.

If I put the effort into setting this up, would anyone (you) use it?  Would you care if I did anonymous data analysis using IPs to track players as long as I didn't post names?  Would you prefer names? 

[Kowal]

I for one would not use it.  Reason being, existing servers already do what I aim to accomplish.  Given moments when I'm inclined to do testing, all I have to do is harass Jacob Orlove, Justin Timoney, Jesus Roxas, Rich Shay, Josh Silvestri, etc.  While not every player has this option available, it does speak volumes for those of us who already do regular testing.

I would anticipate 90% of your positive responses will be from people just outside the, pardon the term, vintage celebrity bracket.  People who know they're decent players, but are not yet ready to compete with the big names either because they haven't put forward enough effort to increase their skill levels, or because there are still mental barriers present when across the table from someone who wrote the article about the deck you netdecked for the event.

The other 10% will be Rich Shay, who gets bored often enough that he'd be willing to play any TMDer over the randomites who he manages to find on mwsplay during weekdays.

[mongrel12]

I'd use it, if it was actually well publicized. I'd suggest talking about it over on the SCG forums, so that people actually use it.

[AmbivalentDuck]

See, the problem with advertising on the SCG forums would be the volume/bandwidth.  If I could keep the usage down to a few hundred active TMDers, I wouldn't have to spend so much money on bandwidth that I'd need to charge/accept donations.

I think it would be cool to have server and the TMD server interact and provide rankings. 

I mean, it'd be nice to know exactly where people stand...and that would become very clear over a few hundred games.

As a question for the site admins: if I just did the software end, would you have the extra bandwidth to host the server?  I would mainly set up a way to integrate the rankings and deck stats into the PHPBB software you're using to run the site.  I believe you auto-log IPs anyways, so keeping track of players on both software sets simultaneously wouldn't be too too hard.

[SpencerForHire]

I could see a server being useful so we don't have to use mwsplay.net if we are unable to host.  For the most part I have no issue finding opponents without a server though.

[AmbivalentDuck]

I don't think the main point would be *finding* games, so much as just keeping track of them.  It would really cool to actually have a breakdown at all times of the top 200 cards in the format, top decks, and top players...with numerical results to back it up.

[Jacob Orlove]

As a question for the site admins: if I just did the software end, would you have the extra bandwidth to host the server?

I'm pretty sure the answer to that is no, but we'll have a better idea once the ad banner goes up, I imagine.

Also, the last time this was tried, not enough people showed up and it died.

[The Atog Lord]

Kowal's right, sometimes I do just go into MWSplay and find a random game.

Duck, this is a great idea, and I'd really like to see it in action. There are, however, two slight slight problems with it.

FIrst, I often use MWS to test crazy, far-out ideas that I know full-well have a very good chance to not working too well. Yes, I wouldn't be testing the ideas if there were no chance they'd work, but I often do explore ideas that have, maybe, a ten percent chance of working. If matches were counted for some tally, then this server wouldn't be a great place to bring unpolished, rough ideas.

Second, more practically. The basic problem with the ideas is this. We want a place where we can go and find a game at 4 am on a Tuesday night. And we also want some degree of exclsuive access. I suspect that having enough people such that there is almost always a Type One game available (as is the case on MWSplay) will also mean that we can't just open the server up to the "good players" from TMD. And, after one or two instances of checking and seeing that noone is online, many players would likely revert to MWSplay. If this could be overcome, the idea would be great. However, I'm not sure how we could solve this problem. Part of being "exclusive" means that you won't always be able to find a game.

[SpencerForHire]

If everyone uses it instead of MWS we may possibly have a chance of growing a player base.  Try it and we'll see if it flops or not.  That is the only way to be sure.

[AmbivalentDuck]

Summary: I could probably host it for a week to grow the player base (starting when I get it up, likely early June).  It would primarily feature a ranking system for ease in selecting opponents and integration with TMD to provide players with rankings on this site.  By July, I would add data analysis for decks.

I'd need to find bandwidth somewhere because I can't keep it small enough to host myself and still have it be desirable.  We have to attract a high volume, but still keep it easy for high level players to find one another.  And we can't penalize testing of new ideas.

Specifics:
-If the idea ends up being good enough, tracking down bandwidth shouldn't be hard and I could host it for the first week (maybe).  This is a problem to examine once I actually have software.

-High volume is trivial once we have bandwidth.  I'd be very surprised to learn that people in the community at large wouldn't be interested in being able to show exactly where they stand at a given point in time.  I'd also be surprised to learn that the community at large wouldn't be interested in seeing which decks are actually the best over a few thousand trials.

--I could outright write a module for netdecking with metaflex slots.  Just let them download pre-made "netdecks" from a server...small downloads no real bandwidth usage.  Then only let them change a small number of predetermined slots for it to count as the original netdeck.  If we wanted to get fancy and someone just had money sitting around to throw at this, we could also use a neural network to categorize decks.  While those aren't too too "heavy" I can't imagine it could run on the same server.

-Just like on the Yahoo Games servers, we could allow players to select if a given game is rated or not.  I'm thinking maybe three categories: Rated, Testing, Casual. 

Questions:
Any input on how to actually rank people? 

How "open" should this be?  Just let people use ratings to select opponents?  Exclusion for vulgarity/poor sportsmanship?

[freakish777]

I think this is a good idea in theory.

What you'd want to do though is write some code for signing people in (to keep track of ratings/rankings and to deter lazy people from signing up in the first place).

I'm not entirely sure you'd really need any money to throw at a Neural Network.  Although getting paid is nice, writing and training a neural net isn't ridiculously hard.

I have said to people before, if WotC had every set on Magic Online, I would stop buying new cards.

Additionally having access to that type of Metagame knowledge would be interesting (and potentially people would pay for it?  To cover bandwidth costs).

[AmbivalentDuck]

Oh, I wouldn't need money to write the neural network.  I'd need money for a cluster to train it.  They usually use clusters to train the networks, even though they run really light once they're working.

[Lou]

Additionally having access to that type of Metagame knowledge would be interesting (and potentially people would pay for it?  To cover bandwidth costs).

Fuck no, no one is paying for that shit.  Why when they can find a game for free?  And you can't just get on MTGO and play for free.  Do you have any idea how much online power and shit like that would cost?  Maybe someone will quote an accurate price of a set of Pernicious Deeds to show you, as I am pretty sure that they are outrageous.

[SpencerForHire]

Additionally having access to that type of Metagame knowledge would be interesting (and potentially people would pay for it?  To cover bandwidth costs).

Fuck no, no one is paying for that shit.  Why when they can find a game for free?  And you can't just get on MTGO and play for free.  Do you have any idea how much online power and shit like that would cost?  Maybe someone will quote an accurate price of a set of Pernicious Deeds to show you, as I am pretty sure that they are outrageous.

I think a more eloquent way of saying this is:

You are not going to get people to pay for something they already get for free.  If you were to put up a server it would be your responsibility to make it worthwile to the community because as a whole the community has not been proven that this is a worthwile expenditure.  You would have to pay server bills etc.  As you can see TMD costs money and we don't have to pay for it.  Whenever a great resource comes out someone usually has to bite the bullet and pay for running it themselves.

[AmbivalentDuck]

Or just force people to view/click banner ads.

Pay would mean like...$1/year...and that would probably be in services rendered (viewing/clicking banner ads), not actual money.  And surely not online cards. Come on, it would be using the MWS engine.  I don't want people to have to download any new software.

Obviously, the neural network would be a long-term goal.  In the short term, people might have the option of turning off banner ads in exchange for archetyping decks or some such so that we could have the very useful composite analyses available.  In order to actually have the neural network, it would likely take 5-10 hours on a cluster once a month to retrain it.

[Evenpence]

I would love to use this.

[freakish777]

Additionally having access to that type of Metagame knowledge would be interesting (and potentially people would pay for it?  To cover bandwidth costs).

Fuck no, no one is paying for that shit.  Why when they can find a game for free?  And you can't just get on MTGO and play for free.  Do you have any idea how much online power and shit like that would cost?  Maybe someone will quote an accurate price of a set of Pernicious Deeds to show you, as I am pretty sure that they are outrageous.

You apparently misunderstood what I was saying.  I was saying that perhaps people would pay for the metagame knowledge specifically that was collected if the server got enough throughput.  Not to play on such a server (as MWS is free).  If you did understand what I wrote, perhaps you could clarify exactly what you mean.

Oh, I wouldn't need money to write the neural network.  I'd need money for a cluster to train it.  They usually use clusters to train the networks, even though they run really light once they're working.

By a cluster, you mean just a bunch of data points (in this case each data point would be a decklist), or the machines you'd want to run the neural net on?


EDIT:  After re-reading you latest post, I'm pretty sure you mean the machines you'd want to run the Neural Net on.  If run time is an issue, why not either use ANFIS or a different approach like Case-Based Reasoning?

[Evenpence]

What's prohibiting us from getting this started and seeing how it goes?  I know I'd be on this almost constantly if only TMDers knew about it.

[AmbivalentDuck]

Free time constraints.  I'm guesstimating about 100 hours of coding/debugging to get a useful beta version up.  If it was smaller, I'd just do it over spring break...but this looks like it will have to be done this summer. 

[Nazdakka]

Free time constraints.  I'm guesstimating about 100 hours of coding/debugging to get a useful beta version up.  If it was smaller, I'd just do it over spring break...but this looks like it will have to be done this summer. 

Why not just get a server up first and worry about the rest later? If you can establish the basic principle of a TMD game server as being workable first, then you'll have a platform to tack all the rest of it onto. The MWShost server software is available at magicworkstation.com, so no programming needed, all you need is a bit of tinkering to make it work and somewhere to host the thing.

[Mantis]

Would definately use it! Great idea and hope you will succeed. We don't all have such influential friends, if you could get this thing started I will use it for sure!

[Jank Golem]

I just gave up on the old magic program and downloaded mws. I would be interested in this and know some other people who would be too. Good Luck!

[AmbivalentDuck]

The problem with just putting the server up and having it censor out non-T1 games is that I'd be eating up my bandwidth with minimal gain from doing so.

[Evenpence]

Let's do this!

[orgcandman]

Something that would be much better would be lexical parsers and interpreters that understood the rules of the game, and how to play the game. Then have a bunch of bots always able to play.

[freakish777]

Something that would be much better would be lexical parsers and interpreters that understood the rules of the game, and how to play the game. Then have a bunch of bots always able to play.

My head is exploding at how terrible of a statement that was.

You do realize just how hard of a problem that is right?

If you were able to code a program that could "teach" a machine how to "read/understand english" you'd be a trillionaire.

Next, we already have the rules of the game coded, just not available to us, in the form of Magic Online.  Getting a program to play Magic is beyond hard. 

Using Machine Learning techniques, the state-space is far, far to vast even if the decks are the same for every match the program plays.   

Using a Heuristic (even something simple like, maximize CA and keep life above opponents life total) in something like expectimax is potentially doable, but there's problems here too.  First, not everyone is going to agree on a heuristic (Magic: the gathering is still in its infancy compared to say Chess which has about a century's worth of theory and testing behind it, we're pretty sure of the heuristics at this point).  Second, any simple heuristic is going to perform poorly.  Third, any complex heuristic is not guarenteed to perform well, but is guarenteed to take lots and lots of time to run through the search tree (which will be fricking huge due to the fact that there will be 10 to 40 branches simply from the unknown of the top of your deck, unless we make it cheat [see all cards and all opponent's cards], in which case we'd just have ABMinimax, assuming no shuffling).  By lots of time I mean hours if not days.  There's more problems here, but like I said, my head is exploding which makes it hard to think.

Using Neural Networks to recognize patterns of play that are advantageous to you (the machine), would take months just to get something that actually plays it's land for the turn (ok, I'm exaggerating this one, but in all seriousness, coding the Neural Net would take a fairly long time, and running it would be insane, Go is arguably an easier game then Magic, here's a paper on making Neural Nets play it:  http://citeseer.ist.psu.edu/donnelly94evolving.html )


Anyways, if you think this is doable, by all means, please code us up some bots, and then write a paper and send it in to AAAI, and go to your local patent office and patent the interpretators.  You'll be rich.

[Danzig]

No Legacy, no care.  ;D

[AmbivalentDuck]

Legacy wouldn't be that different from vintage.  I mean, rankings are rankings.  Depending on traffic, somewhat else would/might have to host it.  Same software, though.

[Godder]

Using a Heuristic (even something simple like, maximize CA and keep life above opponents life total) in something like expectimax is potentially doable, but there's problems here too.  First, not everyone is going to agree on a heuristic (Magic: the gathering is still in its infancy compared to say Chess which has about a century's worth of theory and testing behind it, we're pretty sure of the heuristics at this point).

And the rest - while higher chess theory is constantly being refined, the basics have been in place for centuries, and even basic positional play can be found in Philidor's work (mid-late 18th C).

Using Neural Networks to recognize patterns of play that are advantageous to you (the machine), would take months just to get something that actually plays it's land for the turn (ok, I'm exaggerating this one, but in all seriousness, coding the Neural Net would take a fairly long time, and running it would be insane, Go is arguably an easier game then Magic, here's a paper on making Neural Nets play it:  http://citeseer.ist.psu.edu/donnelly94evolving.html)

Go is easy to get a computer to play, but computers can't play at all well because the tactics are beyond current hardware capabilities. I think Magic would probably fall into the same category - the calculations would be insane. That said, while Magic is tactically complex, there's not terribly much strategy in actually playing a game - most strategy comes during deck selection and construction.

[AmbivalentDuck]

That said, while Magic is tactically complex, there's not terribly much strategy in actually playing a game - most strategy comes during deck selection and construction.

I have to disagree with that.  Deck selection is easy, or at least it could be.  With more numbers to choose based on, an algorithm could easily select the "best deck."

And play is insane: a lot of "advanced" play skill is the ability to read your opponent.  Get an algorithm to do that without cheating and I'll buy the technology for $100M.