So I recently pre-ordered a game called "Defcon". I'd read about it recently and thought it sounded interesting. Pre-orders were only $10! What a deal.

Defcon is ripped straight out of the movie WarGames (which is going right to the top of my Netflix queue, so help me!), and they nailed it on so many levels.

In the movie WarGames, our hero Ferris (whatever) is a high school gamer who can't wait for the next big game to come out. So he attempts to break into the software company (online, not physically) and grab an early copy. He finds some computer system with games on it, and assumes it's the right one, and begins to play "Global Thermonuclear War" ("How about a nice game of chess?" the computer asks. "No... I want to play... global... thermonuclear... war" Ferris responds, speaking as he types. "Fine" says the computer. Classic!).

The game that subsequently begins is the basis for Defcon. A huge world map, drawn in vector graphics, scrollable and scalable. Each player controls one of 6 territories, and begins by placing radar, missile silos, airbases, and naval fleets. And then the Defcon counter begins counting down, from 5 (peace) to 1 (let the nukes fly).

Both games (Defcon and GlobalThermoNuclearWar in Wargames) simulate a full-world nuclear war. Defcon simulates WarGames' primative vector graphics and glowing screen. There is this really creepy mood music playing, and you can't help but imagine that you're in a secured underground bunker somewhere. The tension builds during the game, as players create and break alliances, secretly move hidden subs right off each others coastlines, and eventually all hell breaks loose as Mutuall Assured Destruction comes to pass. When a nuke hits a city there is a white glow, a single, brief piano note stands out from the constant mood music, an almost inaudible rumble, and a message pops up: "London hit, 5.1m dead". The juxtaposition between the massive violence of the game (if measured by "body count") and the graphical quality of that violence (which is to say, completely non-existant) is really moving. It's eerie.

In WarGames, the game that Ferris started running was actually a military simulation, and that simulation begins looking for the winner of a nuclear war, trying tactic after tactic. The simulations run faster and faster, the screen going crazy with flashing lights as the world destroys itself over and over again.

US first strike. Winner: none.
Soviet first strike. Winner: none.
China first strike. Winner: none.
China-US versus Soviet. Winner: none.
China-Soviet versus US. Winner: none.
India (Soviet ally) vs Pakistan (US ally). Winner: none.
Israel vs an Arab state. Winner: none.
... and a hundred more.

Finally, the screen goes dark, and the computer says "A strange game... The only winning move... is not to play."

A Defcon game ends after everyone has shot their nukes off. There are a few different game modes which modify how a winner is determined (either most kills, most survivals, or a hybrid of both). As the game ends, we are left with a mostly obliterated world map, glowing green fallout clouds covering most major population centers, and the message that no one really wins the game as much as everyone loses, but someone loses the least, becomes clear.

I've played games where I've killed zombie demons with chainsaws or gunned down UN security forces trying to defuse the C4 bomb that I planted, and worse... who'd have thought one of the most thought-provoking images of violence would be a single spot of white light on a map?

---

I was listening to the radio on the way to work, and there was a commercial for real estate loans. A while back, they'd advertise how low their rates were, especially getting an adjustable rate mortgage at like 3%. Now, they say "Stuck with an ARM that is going through the roof? Refinance now with our low 30 year fixed..."

Those guys are bastards.

---

For the past few days, I've been obsessively thinking about this Netflix programming contest. The idea is that they want to improve their recommendation system. They are offering $1 million to anyone who can improve their user->movie prediction by over 10%, or $50K a year to anyone who can improve it by 1%. To do this, they provide 100 million data rows looking like (userID, movieID, rating). They also provide a test... that is, about 3 million users and movies, but withhold the ratings. Our job is to try to predict what the users actually rated the movie.


The main strategies appear to be as follows (given user U and movie M)

1) Find other users who have similiar tastes as you, based on ratings over common movies, and find what those users rated M. Take the average of that, and there you go.

In other words, if I trust my boy Iwan (meaning, we share a lot of common ratings), and Iwan tells me he really liked movie M, I will probably like it too.

and/or 2) Find movies that are similiar to M, and find out what users rated those. In other words, if movie M is a lot like movies A, B, C, and D, and I liked all those, I'll probably like this one as well.

Pretty interesting stuff. It's an easy problem to wrap your brain around, but not easy to solve. The first obstacle has been dealing with 100 million data points. The second obstacle will be finding solutions for 3 million rows. If my algorithm takes 1 second per row, that's still over 3 days of processing time. Ugh.

I have a date with a workboy tomorrow to see what we can come up with. I've got some good ideas, and he's a pretty smart guy, so we'll see how it goes.

Ciao.