Use root-sum-square instead of arithmetical mean

[zachanassian]

Brought up here.

At the moment eg. final win% is the arithmetical mean of thw win% of each deck. But what would be the meaning of root-sum-square of the individual win%? Please add some explanation/details

[n00way]

I lost in terms.
That was root-mean-square.

win% = sqrt((win1^2 + win2^2 + ... + winN^2) / N)

[zachanassian]

Calculation is more or less clear but what is the practical meaning of the calculated value.

e.g. win% = (Win1 + Win2 + ... + WinN) / N

practical meaning: is the average win percentage I will have if I fight each deck 1000 times.

But what is the practical meaning of

win% = sqrt((win1^2 + win2^2 + ... + winN^2) / N)?

[n00way]

Sorry again. Root-mean-square is not a solution for my issue. :(

Issue: when I sim against easy and hard decks, app maximize easy decks win% because it's easy and bigger contribution in final score. My idea is to have win% delta different so as contribution of decks with high win% would be lower than contribution of decks with low win%.
Currently don't have solution for implementation.

e.g. start: 10 10 500 900 900 900
kandidate1: 100 100 550 910 910 910 (delta 260) - hard decks improved
kandidate2: 20 20 510 980 980 980 (delta 270) - easy decks improved

kandidate2 wins just because delta is bigger, when kandidate1 has more potential.

[zachanassian]

Got the idea. Thx. Unfortunaltly I am not the author of the climb logic. I just took the source code from WMT and adopted it to TU. So this is out of my league as well. Will keep it here. Maybe somebody has an idea how to implement this.