Stats Discussion

[infinitecat]

It bugs me that in almost all cases on that list node-destruction exceeds node-building by a large margin. How is that even mathematically possible?

Going by the numbers, Zilch and Bisounours are the two players on that list that best exemplify the selfless team-play Onslaught was designed for.

[pooty]

It bugs me that in almost all cases on that list node-destruction exceeds node-building by a large margin. How is that even mathematically possible?

That list is an hourly rate. Shooting/destroing nodes is much easier than building. One good run in the mino on minus might net you 4 or more destroyed... much hard to build 4 nodes.

You don't get any points for defending 1/2 on dria with a shield tank, but there I am helping my team for no score. I've never seen you in a shield tank.

There are valuable team play aspects such as that, that you wont see stats for.... Last night on Torlan-LM, Anon was camped in the Heavy Link Tank, annihilating every flyer/vehicle coming to 2/4 (much to my frustration, its like he knew where I was going to pop up!). Did he get node points no, but he stopped me at least 4 times from destorying that node.....

I say lets be a bit more team like, as a community, there's different play styles and complaining is fine for specific maps/matches ("Hey douche, you really screwed us with your leeroy jenkins mino run) vs. ("Hey you're a shitty player"), one is more personal than the other.

[Judas]

I'll preface this by saying I agree that I should engage people at nodes more and choose faster vehicles if I wish to be a better asset to my team. I always appreciate constructive criticism.

I think win % matters most. Enyo, Anon, leon, and Xexx have the highest win rates, around 60% or higher. I think we all agree that these are the strongest players on the server. The presupposition here is that higher Node Hour Summary values mean that people win more often. I did a couple quick correlation measures on the top 40 players in https://www.omnipotents.com/utstats/onsrankings.php. I used a Spearman's rank correlation. For reference:

0-0.19 is regarded as very weak
0.2-0.39 as weak
0.40-0.59 as moderate
0.6-0.79 as strong
0.8-1 as very strong correlation

For PPH vs win %, I got a correlation of .394 -- weak/moderate.
For Node Hour Summary vs win %, I got a correlation of 0.2 -- very weak/weak

Both of these numbers by themselves aren't great indicators of winners, but Node Hour Summary is significantly less indiciative than PPH. We know that nodes give more points than kills, and nodes are a part of PPH scores. Perhaps we might want to examine how PPH is calculated so that we can use numbers to better inform our assertions in the future.

[pooty]

Interesting. I wonder though if you just took Nodes Constructed or Nodes Destroyed? The Const Nodes Dest / Hour is nodes that are started then destroyed before built... I suspect PPH is probably the strongest out of all the stats...

And lately PPH is most influenced by building a node, destroying a completed node, linking a node....so in a way its dependent on Node stats.

That being said, there's multiple variables (features) that contributes. Even using multiple individual stats don't tell the whole story... those stats don't capture teamwork..... I'll postulate that if you put all the snipers together is worse, than a mix. Likewise with aggressive players. So team composition is key.

I can tell you for example, depending who's on my team I can alter my play style, especially if its an aggressive team, I can focus on more "big" actions (deemers, minos) than building nodes.

[Judas]

Interesting. I wonder though if you just took Nodes Constructed or Nodes Destroyed? The Const Nodes Dest / Hour is nodes that are started then destroyed before built... I suspect PPH is probably the strongest out of all the stats...

Using flat values like in the Power Nodes Constructed or Power Nodes Destroyed which are accumulated over time won't tell us anything when correlated with a percentage, unless I'm misunderstanding something.

That being said, there's multiple variables (features) that contributes. Even using multiple individual stats don't tell the whole story... those stats don't capture teamwork..... I'll postulate that if you put all the snipers together is worse, than a mix. Likewise with aggressive players. So team composition is key.

100% true. I just wanted to make a point here: picking a specific stat value, mapping it to an arbitrary subset of players, and then creating a narrative around that when the numbers themselves tell a different story is problematic.

It'd be a fun task to incorporate a given player's performance in other contexts, such as who else is on a team, the map they're on, time of day, ping, # of beers/joints/coffees consumed, and so on, as additional features in the data. Even qualitative features like roles that you're describing such as 'sniper' or 'defense' might be informative. That might give us a better picture of how the player performs.

Happy to discuss further in the balancer thread -- I don't want to drag this discussion off-topic.

[pooty]

.

[Judas]

Pooty, I can probably do more if I can get per map stats for all the matches we have. What would be the best way for me to do that?

[pooty]

So all the stats are in the Stats DB. Its not open to the internet.
But take a look at the stats pages and let me know what you need.
I can write a custom query to pull all the data needed down and send it/post it on the website, as a csv file.

[Judas]

CSV works fine. I can do any file format -- whatever is easiest for you.

I can start looking match by match, with data like https://www.omnipotents.com/utstats/mat ... match=7520. From there, I can filter out matches with less than 10 players, then filter by map, and see if I can put together anything interesting. My goal is to spruce up the ol' GitHub with a data analysis task.

Thanks for working with me!

[pooty]

File format is easy.

Hard part is the statsDB data model is a freaking mess (its what happens when DIY person builds a database vs. good modelling practices)....