Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Steam Leverages Machine Learning in New Interactive Recommender - MMORPG.com News
SBFord
Former Associate EditorMember LegendaryPosts: 33,129
Steam Leverages Machine Learning in New Interactive Recommender - MMORPG.com News
Steam introduced a new "interactive recommender" today, dubbed, um, Interactive Recommender. According to Valve, they've apparently heard our cries for better tools to help us discover games specifically tailored for the individual player.
Comments
Sometimes I feel like one of those action heroes who manages to catch on to the back of the moving train car but is hanging on for dear life with their legs dangling out in thin air.....
With a grinning millennial trying to pry my hands off of the railing while saying, "let me help you out there."
Young bastards.
TGIF
"True friends stab you in the front." | Oscar Wilde
"I need to finish" - Christian Wolff: The Accountant
Just trying to live long enough to play a new, released MMORPG, playing New Worlds atm
Fools find no pleasure in understanding but delight in airing their own opinions. Pvbs 18:2, NIV
Don't just play games, inhabit virtual worlds™
"This is the most intelligent, well qualified and articulate response to a post I have ever seen on these forums. It's a shame most people here won't have the attention span to read past the second line." - Anon
I need games with AI which would learn my gameplay for few hours so I could let it play for me and watch the gameplay instead. This way it'll be easier to see my mistakes + I don't really enjoy mouse clicking and pressing buttons so watching games is going to be my primary "gameplay" in the nearest future. Also it allow to spend time doing something useful in parallel while nothing interesting is happening on screen - like learning new languages.
"AI" players are already implemented in Mortal Kombat 11 and it's wonderful. Need more games like that.
Thank you for your time!
I just did an awesome so when AI takes over they know I was part of their fifth column.
Aloha Mr Hand !
Vault-Tec analysts have concluded that the odds of worldwide nuclear armaggeddon this decade are 17,143,762... to 1.
“Microtransactions? In a single player role-playing game? Are you nuts?”
― CD PROJEKT RED
Logic, my dear, merely enables one to be wrong with great authority.
Fortunately for me they aren't very adept at understanding sarcasm, snarkiness and the overuse of hyperbole so I'm still pretty safe.
"True friends stab you in the front." | Oscar Wilde
"I need to finish" - Christian Wolff: The Accountant
Just trying to live long enough to play a new, released MMORPG, playing New Worlds atm
Fools find no pleasure in understanding but delight in airing their own opinions. Pvbs 18:2, NIV
Don't just play games, inhabit virtual worlds™
"This is the most intelligent, well qualified and articulate response to a post I have ever seen on these forums. It's a shame most people here won't have the attention span to read past the second line." - Anon
Seems fair.
"True friends stab you in the front." | Oscar Wilde
"I need to finish" - Christian Wolff: The Accountant
Just trying to live long enough to play a new, released MMORPG, playing New Worlds atm
Fools find no pleasure in understanding but delight in airing their own opinions. Pvbs 18:2, NIV
Don't just play games, inhabit virtual worlds™
"This is the most intelligent, well qualified and articulate response to a post I have ever seen on these forums. It's a shame most people here won't have the attention span to read past the second line." - Anon
P.S a third of the choices for me were in their 60's in the end, if it does not factor that it is not of much use.
거북이는 목을 내밀 때 안 움직입니다
"True friends stab you in the front." | Oscar Wilde
"I need to finish" - Christian Wolff: The Accountant
Just trying to live long enough to play a new, released MMORPG, playing New Worlds atm
Fools find no pleasure in understanding but delight in airing their own opinions. Pvbs 18:2, NIV
Don't just play games, inhabit virtual worlds™
"This is the most intelligent, well qualified and articulate response to a post I have ever seen on these forums. It's a shame most people here won't have the attention span to read past the second line." - Anon
Aloha Mr Hand !
The precursor to machine learning is the ordinary least squares regression. Or more colloquially, fitting a line to your data. Given a bunch of points in a two-dimensional space, there is some unique line that best fits the data in some technical sense. A lot of you have probably done this at some point. When you have more variables, it's geometrically fitting a hyperplane to your data in some higher dimensional space, but the math is nearly the same.
The basic idea of machine learning is that rather than having to fit a line to data, it can try more complicated equations. Some types of machine learning aren't that computationally expensive, but some are, especially the neural networks that get so much attention from hardware vendors precisely because they take a lot of hardware.
The more things you have to try, the more computationally expensive it is. Neural networks try an enormous number of things, as do random forests. Loosely, they try a ton of things in some reasonable manner, and report back that this is the equation that best fits your data.
You can look at the output from some machine learning thing and say, where did that come from? The computations are too complicated for anyone to understand all of the details of exactly where the answer came from, even if some people do understand the formulas pretty well.
But that the computations are so expensive isn't new. Ordinary least squares regressions have that same feature in all but the simplest of cases. That the computations are so heavy is why only the simplest of cases could be done by hand before the advent of computers.
Here, understanding what the computations need to do at a high level isn't that hard. It's just some matrix multiplication, plus taking the inverse of one matrix. But you're not going to take the inverse of a 20x20 matrix of arbitrary numbers by hand, or at least not with sufficient precision to avoid drowning in rounding errors. You're just not. People can look at the output and say, that looks reasonable, especially when it's just fitting a line to some two-dimensional data. But you don't really understand exactly how those coefficients came to be.
There are really two major types of machine learning: supervised and unsupervised. The former is closer to what I described above. You have some data, then run a machine learning algorithm to find formulas to fit the data. Then given those formulas, you can plug in other data points with your dependent variable missing and have it approximate the dependent variable as best as it can.
The idea of unsupervised machine learning is that you have a bunch of data that you'd like to know something about. You run it through the algorithms and it partitions your data into clusters of similar data points. For example, if you had the latitude and longitude of the home address of everyone in a country (or some other region), unsupervised machine learning could tell you where the big cities are.
Or, to return to the topic of this thread, you can recognize that some games are similar to each other, or that some players tend to like similar sets of games. By keeping track of which sets of games tend to be played by the same players (e.g., if a player plays game A, he's far more likely to also play game B than one who doesn't play game A), you can identify genres--and put new games into genres. And once you see some games that a player likes, you can identify the genres that he likes.
But while machine learning can do a lot, it's not the solution to everything. For starters, some problems are easy to solve without it. My example above of finding where the big cities are is a case of this. You could do it with machine learning, but that's a lot more work than doing the same thing by other means.
Some problems just aren't solvable by machine learning. If what you'd like to know just isn't in your data, then machine learning can't help you. I once heard an example of a shipping company that knows the size and weight of its packages, where and when it was sent from, and where it is scheduled to be delivered to. And what they wanted to know is which packages contain bombs or other such contraband.
Where machine learning does best is when you have a lot of data and it seems like the answers you're looking for are in the data you have, and some formula applied to that data will give you the answer you want, but it's not at all obvious what that formula should be.
Another important point is that machine learning really only works on problems where it's okay to be wrong a lot. If there are ten games on Steam that you'd like but have never heard of, and a machine learning recommender can pick out 20 games that includes 5 of the ones that you'd liked to have known about, that's valuable. The gamer can manually look through a list of twenty far more easily than a list of thousands.
If out of 1000 murders, a machine learning algorithm can pick out 2000 suspects, 200 of which are actually guilty, but you have no idea which ones, that's not helpful. Or at least, it's not helpful if you put a high value on not convicting innocent people. If you're an evil dictator trying to find who is plotting against you and are willing to execute 10 random, innocent people for every dissident you catch, maybe you go for it.
If you find that all baffling, then this might be comforting:
That is actually, technically true, by the way.