Simple Gamescore for Dota 2
When I first started watching Dota one of the most confusing things was trying to figure out who had the advantage. One would assume, at first, that kill count is the end all be all. However; That assumption quickly falls apart once you see a team with fewer kills convincingly win a game. So, to address this question I turned to 2 of my favorite past times: chess and baseball.
In chess the use of computers to assess the state of the game is now the norm. The computers are great at brute force calculation, but programmers also direct the computer by giving certain positional situations a - or + value. The read out and scaling of the engines is what we are going to use. In a chess game a score of 1 would mean the white is 1 pawn ahead of black. A score of -3 would mean that black is 3 pawns ahead. We will use this same idea, with positive numbers being the radiant and negative being the dire. In chess the scale is by pawns, in Dota we will use kills.
From baseball we will take the titular name for this stat. Gamescore was a statistic developed by Bill James to evaluate how well a starting pitcher pitched in a particular game. He gives value to almost everything that a pitcher does. For instance: a strikeout is one point, an out is one point, giving up a hit is -2 points etc. This is a really wonderful idea for many reasons, but one of my favorite things about it is how simple it is to keep track of. I can quite easily keep track of this statistic in my head while watching a game live, since it is just simple addition and subtraction.
So! The parameters I set myself for this statistic were: be simple to keep track of, have an output that describes the state of the game that is simple to understand. Now, before we go to the nitty gritty lets have a few caveats. This isn’t perfect, it is a very simplistic take on the game. It doesn't take into account many things that affect the game. This formula was something that took me a long time to come to, for awhile I had many more categories and the numbers were more confusing. For example, how do you quantify a courier kill? It is difficult to do and it depends many things; when in the game it happens, what items were on the courier, who killed it etc etc. I decided that questions like this were above the scope of a simple stat like this. In a future blog post I will detail a much more complex algorithm that will deal with questions like those and be more like a chess engine readout. For now lets get to Simple Gamescore.
Here are all of the values you need to know:
Kill
|
1
|
Tier one tower
|
1
|
Tier two tower
|
2
|
Tier three tower
|
3
|
Tier four tower
|
5
|
Rax
|
4
|
Roshan Kill
|
1
|
Aegis
|
1
|
Cheese
|
1
|
That’s it, simply add up the scores for both sides that subtract radiant-dire.
The first test for this is to see if the team that wins always has the higher gamescore. This can be a little tricky due to teams being able to call gg whenever they want. You can imagine a team throwing a game and calling gg before the gamescore has time to catch up. Even in this case the score should get at least close before the game ends.
Lets talk about range, how much of an advantage in gamescore does a team need to win? Here are the very approximate numbers that I’ve come up with from running gamescore on about a hundred games.
Points ahead
|
Advantage
|
0 to 5
|
Even
|
6 to 10
|
Slight advantage
|
11 to 15
|
Large advantage
|
16 to 20
|
Huge advantage
|
20 and up
|
Game over
|
One more thing I should say about this is as the game goes longer the score needs to be higher for the team to win. A 10 point lead in the first 10 minutes means a lot more than a 10 point lead 50 minutes in.
To begin our examples here are the results of a few recent games. In the interest of not driving the author crazy, these are not perfect. DatDota does not provide details on Roshan kills. Even with that part of the equation left out you will see that Gamescore still shows the winners, which is the real point of this section.
Match ID
|
Teams
|
Final Gamescore
|
Result
|
564209414
|
C9 - Liquid
|
-51
|
Dire
|
561178523
|
EG - For Sweet
|
32
|
Radiant
|
562305704
|
Sigma - Empire
|
-31
|
Dire
|
563999328
|
EG - VG
|
22
|
Radiant
|
Of course this start isn’t as interesting after the game, it is during the game that we want to know the information. Here are a few charts of the final game of TI3.
The first chart is simply the total game score for each team during the game. The second is the radiant - dire score which I mentioned earlier. In a perfect world I would have a program that could record the gamescore at each moment in the game and then translate that to a graph. Since I don’t have that I took the score from each minute in the game. This, sadly does not allow the graph to capture the effect of the ‘million dollar dream-coil’. The minute that the dream-coil happened was directly after the last blue spike you see on the graph. What the graph doesn't show is that Na’vi had grown their gamescore lead to 10 points in that minute, and the dream coil alone brought it back down to 4 in a matter of seconds. The first big jump of red is when the Rax go down, quickly followed by the next big jump which was the tier 4 towers. In the middle of the graph you can see the small advantage Na’vi had built, the steep drop of the blue there is the s4 aegis steal.
Hopefully this shows how effectively this simple system can show viewers of the game what is going on. It is also useful in showing where the most important moments of the game happen. This is helpful not only to new players who might not know what is going on, but could allow casters to easily find the exact moments where they should show a replay after the game. I hope you find this statistic as useful as I do! Next time we will talk about the theory of making a much more complicated version of gamescore, that could be even more useful to the wider community.
