Stats Discussion

[Judas]

If we can add some extra columns to the spreadsheet for qualitative player features, I can make a little model that will use those + relevant stats to predict an outcome between groups and then try to balance matches around that. Nobody needs to install anything -- I can just plop in some player names for each side and it will do the rest. We'd need to be super careful about adding bias or the model won't be effective. For getting map match outcomes, I can use a scraper and wrangle the data out of there if using the DB is too much of a pain.

Column ideas: wastes important vehicles, aggressive on nodes, DM proficient, vehicle proficient
I can use a 1-5 scale or just boolean as values.

It might also be helpful to have an indicator column of whether a given map is more DM-focused, more vehicle-focused, or both as an additional context feature. DM players do win more often on DM maps like Tyrant or Master Bath, right?

[pooty]

Okay since Tuesday last week.
The KnowPlayerPPH was set to what's in https://docs.google.com/spreadsheets/d/ ... =671937450

Now its not for all the players, but most of the top, and bottom tiers. But it was those ranges of values instead of the "gut feeling"

(Its done by guid, so I have to look up player names in another file, hand enter it into that .ini file
Its sort of begging me to write a python script to update it... )

So how's the balance been over the last 3 or 4 days or so?

[pooty]

If we can add some extra columns to the spreadsheet for qualitative player features, I can make a little model that will use those + relevant stats to predict an outcome between groups and then try to balance matches around that. Nobody needs to install anything -- I can just plop in some player names for each side and it will do the rest. We'd need to be super careful about adding bias or the model won't be effective. For getting map match outcomes, I can use a scraper and wrangle the data out of there if using the DB is too much of a pain.

Column ideas: wastes important vehicles, aggressive on nodes, DM proficient, vehicle proficient
I can use a 1-5 scale or just boolean as values.

PM me your email if you want access to edit the sheet. I was going to poke at it a bit today, but wife decided I need to paint....

I do think it'd be interesting to create a bunch of player features and see how those correlate. I am not sure how we capture player interaction, so it'd be interesting to see from the match data, if say Enyo/Anon are on same team Win PCT is xx % vs. say Enyo/Leon etc.

Judas, if there's a relevant DS/ML model that you think will capture that let me know... If we need some compute/framework we could use https://community.cloud.databricks.com/login.html (I think our data is under 2GB limit) so we could parallelize a bunch of runs. If we need more compute than that...I can probably figure something out.

[Judas]

Thanks pooty. I'm going to start with some older nonsense on sklearn that I used in my MA thesis (https://scikit-learn.org/stable/modules ... ssion.html) and work my way up to more modern stuff. I'll work with a copy of that Google spreadsheet you linked to until I have something substantial to contribute. I don't think we need a huge amount of compute given that I can put anything excessive onto GPU, but I'll let you know.

[pooty]

So we used the newer multipliers for about a week? Thoughts on whether they were better/worse than my hand generated/gut feel ones? (No ego here, so don't worry if you want to say they sucked before).

Only downside here, is I've seen combos of like Anon,Rhamp, SubZero ( I think that was on Bitchslap, yikes), get all put on same team vs being split previously.

And for the masses, my theory is the vehicle wasters, you know who they/you are, hurt the team more than even the bestest players can help.

[Enyo]

So we used the newer multipliers for about a week? Thoughts on whether they were better/worse than my hand generated/gut feel ones? (No ego here, so don't worry if you want to say they sucked before).

I dunno, seems about the same to me. Sometimes teams are balanced, sometimes they’re horribly not so. That seems to be regardless of old/new multipliers or no multipliers at all.

[DrScrotey]

All I know is the last two play sessions I've lost 14 out of 15 games. Maybe my multiplier is too high? There's probably a minimum win percent someone can stand before just moving on.

[pooty]

Maybe my multiplier is too high? There's probably a minimum win percent someone can stand before just moving on.

Its lower now than before....
Right now the spread is much, much less than it was previously. Before they multipliers were higher to basically keep the tier 1 players always seperated out.. now that's not as likely.

I personally feel it was better with the older multipliers...

But at times you do get streaks where you just get repeated bad teams...been there.

[captainsnarf]

I still think the new multipliers are wrong. The formula is wrong.

It should be

Code: Select all

Multiplier formula = 1.0 + (((WinPct/100)-0.5) * scale)

The current values with 2.2 scale is giving players with < 50% win rate a positive multiplier. There is no way that is correct regardless of how good their aim is.

[pooty]

So that formula (scale =1 ) in col H in the sheet: https://docs.google.com/spreadsheets/d/ ... =671937450

Leon gets < 1, because of WinRate is low, which doesn't seem right either, sure it will boost his win rate, but it will unbalance teams to do so.