Data driven player stats

[SuperVehicle001]

I'm a programmer/data analyst. I started thinking, "what if I took a money ball approach to evaluating top level players?"

I'm a numbers nerd. With some help we may gain insight into the meta game at high level play. This is not an attempt to create a tier list of characters or say "who is the best player". It's mainly for fun (at least for me)

Let's invent Kombatball! Or maybe Kombatalytics.

I want to answer questions like, at what point is continuing a combo not effective? How long can a combo go before a breaker given the target has no meter at the start?

At all major tournaments what is the proportion of characters used? Do hypothetical match up numbers match statistically to real data? I could use data analysis to asses accuracy of match up numbers.

At what percentage of health remaining do most players let a combo finish in order to carry meter to the second round? Does letting a combo finish to carry over meter justify risk/reward?

These are just a few questions I thought of.

What kind of questions do you think would give insight into high level play?

The next step for me to take will be thinking of ways to measure and collect the data related to the question. I'll focus on that with a later discussion.

Sent from my Nexus 4 using Tapatalk 2

 

[Momentum]

Great post! I was wondering about this for the sake of my own play.
As a kabal player I would love to know the most common reactions (provided its the first time that string has being used in the game so nobody has been conditioned to give a certain reaction) to;

F3 NDC
F32 NDC
B1NDC
B12NDC
B121NDC
B2NDC
VS every specific character

I've always believed that math=power in games and although it'd be a mind-numbing task to find the most probable option for every situation, I think it would definitely be worth it

Edit: Oh and I've got an infinite amount of scenarios that I'd like to know the statistics on so if you do follow up with this project let me know!

 

[SuperVehicle001]

You are digging deep right now. This will be hard! I wouldn't have it any other way!

Before I start brainstorming a statistical method to use how would we go about collecting data?

My first idea is watch A LOT of recorded streams or tournament videos. It would be tedious and time consuming.

I don't have the knowledge of Kabal to recognize those strings.

Maybe we can work together and get some preliminary data from a few match ups to test out some of my stats ideas.

I'll create a Google docs spreadsheet to capture and share the raw data.

Once we have that we can do some interesting things.

Sent from my Nexus 4 using Tapatalk 2

 

[Mr. Mileena]

Nice.

For example, if I know I'm going to lose the round, I fan lift spam with Kitana to gain meter.
If I see the opponent is close to break, 21 lift, instead of 21 ex fan is what I do.

Also, if they are almost dead Mid combo, i try and finish them off with fanlifts + fans

Like 2,1 lift, instead of doing NJP into B2 or 4, Ill do 21 lift, INTO ground fan, air fan, ground fan, air fan

5 special moves in a row that give her meter.
Where as doing NJP b2 would kill them without gaining any meter yourself.

<3

 

[ryublaze]

Why do players end rounds with uppercut when that's just giving the opponent tons of meter?

 

[SuperVehicle001]

Why do players end rounds with uppercut when that's just giving the opponent tons of meter?

If serious question:
Answering that would require a different kind of analysis. That branches into decision theory, utility theory and risk analysis. You would have to quantify the amount of meter gained by the opponent and weight that number by perceived risk (subjective).
http://en.wikipedia.org/wiki/Utility
http://en.wikipedia.org/wiki/Risk_measure

If rhetorical question:
They suck.

 

[SuperVehicle001]

Great post! I was wondering about this for the sake of my own play.
As a kabal player I would love to know the most common reactions (provided its the first time that string has being used in the game so nobody has been conditioned to give a certain reaction) to;

F3 NDC
F32 NDC
B1NDC
B12NDC
B121NDC
B2NDC
VS every specific character

I've always believed that math=power in games and although it'd be a mind-numbing task to find the most probable option for every situation, I think it would definitely be worth it

Edit: Oh and I've got an infinite amount of scenarios that I'd like to know the statistics on so if you do follow up with this project let me know!

I'm starting with Kenshi instead of going down the list alphabetical. He seems to be popular so I think finding footage will be easier.

My question to you is:
What are the possible reactions to your listed combo strings for Kenshi?

 

[CrimsonShadow]

Hey,

The hardest part of this is just collecting a good sample of data. For every match that happens on stream at a typical tournament, there are one-to-several matches that happen off stream. So you end up will a small representation of what actually went down, and that makes it tough to draw truly representative data from video alone.

It can, however, help you evaluate what happened to specific players in specific instances. It's a lot of work, but it could help more clearly explain reasons behind certain wins and losses. To represent characters or overall top player trends though, you would want a LOT of data.

For example the MKi rankings are based on over 1700 matches, with very simple metrics (win/loss, opponent rating) and I'd *still* like to have a lot more data from certain players. I'll be collecting data all year long, from just about every major tournament, and each match counts toward approxmating some semblance of accuracy.

So many players, styles, and situations.

 

[Dramatica]

I have similar interests and think of these things constantly. Talking about it is interesting and informative, but actually doing it I fear would be prohibitively time-consuming. We need the game to collect the data for us. Imagine if the details of every match ever played were logged and put into some form of data that we could analyze numerically... Oh the possibilities.

 

[CrimsonShadow]

We need the game to collect the data for us. Imagine if the details of every match ever played were logged and put into some form of data that we could analyze numerically... Oh the possibilities.

That' s what the World Tekken Federation for TTT2 is attempting to do:

http://wtf.tekken.com/playercard/view/ps3/fire11smk

.. but I'm not sure how extensive the data is.

 

[SuperVehicle001]

Hey,

The hardest part of this is just collecting a good sample of data. For every match that happens on stream at a typical tournament, there are one-to-several matches that happen off stream. So you end up will a small representation of what actually went down, and that makes it tough to draw truly representative data from video alone.

It can, however, help you evaluate what happened to specific players in specific instances. It's a lot of work, but it could help more clearly explain reasons behind certain wins and losses. To represent characters or overall top player trends though, you would want a LOT of data.

For example the MKi rankings are based on over 1700 matches, with very simple metrics (win/loss, opponent rating) and I'd *still* like to have a lot more data from certain players. I'll be collecting data all year long, from just about every major tournament, and each match counts toward approxmating some semblance of accuracy.

So many players, styles, and situations.

Yeah, I love football stats but they have an army of people to rewatch game footage.

 

[Mr. Mileena]

Why do players end rounds with uppercut when that's just giving the opponent tons of meter?

If they are near a breaker, I can see why, so 2 or 3 more hits are avoided (thus a breaker avoided), and end the round.

It's also safer to land sometimes to avoid dropping a combo.

 

[ryublaze]

If they are near a breaker, I can see why, so 2 or 3 more hits are avoided (thus a breaker avoided), and end the round.

It's also safer to land sometimes to avoid dropping a combo.

yea that's what I was thinking but for some characters you can end with a special easily like Sub-zero can slide which does pretty good damage by itself and say if I'm Scorpion and i caught my opponent with a spear. instead of uppercut you can f4 takedown to build meter and takedown is unbreakable.

 

[Mr. Mileena]

yea that's what I was thinking but for some characters you can end with a special easily like Sub-zero can slide which does pretty good damage by itself and say if I'm Scorpion and i caught my opponent with a spear. instead of uppercut you can f4 takedown to build meter and takedown is unbreakable.

The only reason I can come up with is to perhaps maybe keep them far away at the start of the round? I think slide might keep them too close