Are you an expert with data, have game design sense and want to join Riot?

First Riot Post
Comment below rating threshold, click here to show it.

archangelfighter

Senior Member

01-15-2014

Quote:
Originally Posted by Reiver View Post
I'm a Physics undergrad currently, but I might get back to you in a couple years. High-energy physics is a wonderful exercise in data mining.
Mah ****a. Good luck with the undergrad, it only gets better from there.(And funner.)


Comment below rating threshold, click here to show it.

pro4never

This user has referred a friend to League of Legends, click for more information

Senior Member

01-15-2014

Bah every time I see posts like this it makes me want to continue developing my programming skillz.

@Lyte - I hate to throw this off topic but I've always been curious... what type of job would someone with a background in program but emphasis on feature design and integration be looking at?... Cause I've avoided 'programming' jobs because I just don't feel I'd make a competitive candidate (plus the general career trajectory of a programmer in the gaming industry is horrifying) and so instead have focused more along designing and integrating features

EG: I've done a lot of work writing pservers before but my focus has been on taking the concept of what the game should be and turning it into tangible features and systems that the original game never had.

Anyways, best of luck to all the applicants!


Comment below rating threshold, click here to show it.

MP44 Rommel

Member

01-15-2014

Who wants to work For Riot, all they will do is force you to make boring ****


Comment below rating threshold, click here to show it.

ChaosLogan

Member

01-15-2014

Working for Riot would be awesome on a resume, would strengthen your team-working, design and development, statistical analysis, programming, and pattern recognition skills. All of those are transferable to other careers, should you choose to move on after working at Riot.

If you are lucky enough to get the job in the first place.


Comment below rating threshold, click here to show it.

Lyte

This user has referred a friend to League of Legends, click for more information

Lead Social Systems Designer

Follow RiotLyte on Twitter

01-15-2014
21 of 41 Riot Posts

Quote:
Originally Posted by FrozenXylaphone View Post
How important is the statistical know how?

For instance, I am a computer scientist with advanced sql skills used to create programs that keep track of financial data for my company.
But I have never worked with R or Matlab. I did take statistics in college, but my degree was computer science so it wasn't the focus.
It's critical for this position because we need someone who can break down a problem space or question, determine the right techniques or methodologies to employ, then actually run the analyses themselves and generate the reports in plain English.


Comment below rating threshold, click here to show it.

ChaosLogan

Member

01-15-2014

Quote:
Originally Posted by Lyte View Post
It's critical for this position because we need someone who can break down a problem space or question, determine the right techniques or methodologies to employ, then actually run the analyses themselves and generate the reports in plain English.
Can you give a sample type question that someone might face in this position? Doesn't need to be Riot centric for the example, just a type of problem they will be facing.

To be honest, this has gotten my attention to see if I could do something like this but I am afraid of what I don't know I don't know.


Comment below rating threshold, click here to show it.

Fang0r

Senior Member

01-15-2014

@Lyte, I'm actually running a data mining operation on tribunal cases currently. I started from case 6690000 so I have about a month's worth of data (~17000 cases right now)

Hopefully, I'll get some interesting results for you to look at over the weekend :O (though I wouldn't be surprised if you already have a framework set up for looking at tribunal data >_<)


Comment below rating threshold, click here to show it.

Goumindong

This user has referred a friend to League of Legends, click for more information

Senior Member

01-15-2014

Quote:
Originally Posted by ChaosLogan View Post
Can you give a sample type question that someone might face in this position? Doesn't need to be Riot centric for the example, just a type of problem they will be facing.

To be honest, this has gotten my attention to see if I could do something like this but I am afraid of what I don't know I don't know.
Here is my guess at what it is they're doing.

Suppose for a second you wanted to look at the effect of toxicity on the outcome of a game. You face a problem: Toxicity is caused by negative outcomes in a game. So if you do a simple regression you will probably find a result, but it might not actually be helpful even if it does show what you would expect. I.E. more toxicity = more losses. But since more losses = more toxicity this is unsurprising.

So what you might do is look at how the game plays out. Recognize that gold advantage often predicts the winner of a game and also that gold advantage is a function of play and play is a function of toxicity. You would recognize that this can be modeled by a certain form of time series regression called a Vector Auto Regression. You would know the limits and caveats involved in using a VAR so that you could formulate a way that the data should look in order to perform the analysis.

In this case we would be comparing the change in gold difference across teams and the difference in "negative sayings" across teams(or just negative sayings either one might have issues). We would select data from a game every 1 minute or 30 seconds to create a time series for a single game. We would run the VAR and would get data for the effects of negativity now on the change in gold difference in the future, the effects of the change of gold difference now on negativity in the future, the effects of negativity now on negativity in the future and the effects of change in gold difference now on the change in old difference in the future for a number of different lags (1 minute change vs 2 minute change etc). If we sum these up we can get a semblance of the overall effects for that specific game. If we took a sample of games from random players we could get some sort of "base toxicity" and "base snowball" levels by averaging the coefficient values.

If we took a sample of games from a single individual we could get an estimate of that players toxicity by averaging the coefficient values. This might give us insight into how different players deal with toxicity, how they're effected, why they're toxic etc. It might direct us for different types of players to look at. At the very least doing this on a large scale can create an entirely new data set which we can do other analysis on. (For instance you could check to see if players who were not as effected by toxic players were less or more likely to cause toxicity themselves)

Its my understanding that the person they're looking for would be able to identify the form of the data necessary to answer the question, pull data off the database to create the datasets necessary, run the analysis, then present the data in a a manner that the design team could understand.


Comment below rating threshold, click here to show it.

Clock Code

Senior Member

01-15-2014

I'm watching that talk you did.

Really fascinating stuff.


Comment below rating threshold, click here to show it.

ChaosLogan

Member

01-15-2014

Quote:
Originally Posted by Goumindong View Post
Here is my guess at what it is they're doing.

Suppose for a second you wanted to look at the effect of toxicity on the outcome of a game. You face a problem: Toxicity is caused by negative outcomes in a game. So if you do a simple regression you will probably find a result, but it might not actually be helpful even if it does show what you would expect. I.E. more toxicity = more losses. But since more losses = more toxicity this is unsurprising.

So what you might do is look at how the game plays out. Recognize that gold advantage often predicts the winner of a game and also that gold advantage is a function of play and play is a function of toxicity. You would recognize that this can be modeled by a certain form of time series regression called a Vector Auto Regression. You would know the limits and caveats involved in using a VAR so that you could formulate a way that the data should look in order to perform the analysis.

In this case we would be comparing the change in gold difference across teams and the difference in "negative sayings" across teams(or just negative sayings either one might have issues). We would select data from a game every 1 minute or 30 seconds to create a time series for a single game. We would run the VAR and would get data for the effects of negativity now on the change in gold difference in the future, the effects of the change of gold difference now on negativity in the future, the effects of negativity now on negativity in the future and the effects of change in gold difference now on the change in old difference in the future for a number of different lags (1 minute change vs 2 minute change etc). If we sum these up we can get a semblance of the overall effects for that specific game. If we took a sample of games from random players we could get some sort of "base toxicity" and "base snowball" levels by averaging the coefficient values.

If we took a sample of games from a single individual we could get an estimate of that players toxicity by averaging the coefficient values. This might give us insight into how different players deal with toxicity, how they're effected, why they're toxic etc. It might direct us for different types of players to look at. At the very least doing this on a large scale can create an entirely new data set which we can do other analysis on. (For instance you could check to see if players who were not as effected by toxic players were less or more likely to cause toxicity themselves)

Its my understanding that the person they're looking for would be able to identify the form of the data necessary to answer the question, pull data off the database to create the datasets necessary, run the analysis, then present the data in a a manner that the design team could understand.
Thank you for that explanation.

While the drill down into a dataset and math computations you describe do not scare me (I feel my skill set is robust enough in this area), my fear of not having enough statistical knowledge (a VAR as a concept was new to me) and of not knowing the software tools they use (R, Hadoop, Vertica) make me worried for any attempt I might make.

I can run circles around everyone at my job in Excel because of my strong math and coding background, but I am beginning to feel limited on my resume when I read job descriptions like this.