Upvote Upvoted 4 Downvote Downvoted
TF2 Statistics
1
#1
3 Frags +

In non-update related news, I want to do some statistical analysis of games. Imagine .ss, but instead of only focusing on one match, it could analyze players, teams, matches, seasons, maps, etc.

However, all of that automation is a long way away.

Since I'm incredibly uncreative, I want to see if anyone has any questions that could be answered with statistics. Questions that begin "is there a relationship between..." or "Does A affect B more than C..." etc... you get the point. I can't come up with any ideas, so please let me know what questions you'd like answered.

In non-update related news, I want to do some statistical analysis of games. Imagine .ss, but instead of only focusing on one match, it could analyze players, teams, matches, seasons, maps, etc.

However, all of that automation is a long way away.

Since I'm incredibly uncreative, I want to see if anyone has any questions that could be answered with statistics. Questions that begin "is there a relationship between..." or "Does A affect B more than C..." etc... you get the point. I can't come up with any ideas, so please let me know what questions you'd like answered.
2
#2
2 Frags +

Relationships between damage done / kills gotten on classes, just interesting ratios like heals to ubers or damage taken to deaths.

Like, just looking at stats for each player across the whole season is interesting, you could graph out performance levels and see what teams are strong on x maps or could be like how effective is the clockwork factor or something. Stats are usually interesting unless it's something that no one cares about, like the amount of times the esea ads play.

Relationships between damage done / kills gotten on classes, just interesting ratios like heals to ubers or damage taken to deaths.

Like, just looking at stats for each player across the whole season is interesting, you could graph out performance levels and see what teams are strong on x maps or could be like how effective is the clockwork factor or something. Stats are usually interesting unless it's something that no one cares about, like the amount of times the esea ads play.
3
#3
9 Frags +

this can be tested as long as copy / pasting stats from esea's site works. any specific questions you had in mind OP? I WILL SEE WHAT I CAN DO.

edit : I TRIED SOME STUFF.

FUN DISCLAIMER: I'M NOT A STATS GUY, ALL OF THIS IS PROBABLY VERY WRONG. ANYONE WHO IS A STATS GUY PLEASE CORRECT / TEACH ME. IF YOU DON'T GET THE PICTURE I'M SORRY, I'm BAD AT EXPLAINING STUFF.

http://i.imgur.com/LjhIFCB.png

I regressed some data from SEASON 13's TRIHARDS matches. The stats in question are differences between Tri Hards and their opponents DAMAGE per match, along with differences in UBERS and UBERDROPS.

Basically what this is says is that is ubers and drops and damage are identical between the two teams, Trihards have about a 55% chance of winning (cons_). For every single point of damage Trihards inflict more than their opponent, they gain an extra 0.0078% chance of winning. For every uber more than the enemy, Trihards gain an extra 0.035% chance of winning. For every drop more than the enemy, their odds DIMINISH by a huge 7.06%. Of course, the values are reversed for every unit less than the enemy.

Also notice that the t-scores of the ubers / drops are fairly small, indicating they're not very relevant (although dropping much moreso that ubering.,) while damage is very significant.

Again, this is probably very wrong, anyone who knows more about this that I do please feel free to chime in.

TL:DR - read the post sucka.

this can be tested as long as copy / pasting stats from esea's site works. any specific questions you had in mind OP? I WILL SEE WHAT I CAN DO.

edit : I TRIED SOME STUFF.

[b]FUN DISCLAIMER[/b]: I'M NOT A STATS GUY, ALL OF THIS IS PROBABLY VERY WRONG. ANYONE WHO IS A STATS GUY PLEASE CORRECT / TEACH ME. IF YOU DON'T GET THE PICTURE I'M SORRY, I'm BAD AT EXPLAINING STUFF.

[img]http://i.imgur.com/LjhIFCB.png[/img]

I regressed some data from SEASON 13's TRIHARDS matches. The stats in question are differences between Tri Hards and their opponents DAMAGE per match, along with differences in UBERS and UBERDROPS.

Basically what this is says is that is ubers and drops and damage are identical between the two teams, Trihards have about a 55% chance of winning (cons_). For every single point of damage Trihards inflict more than their opponent, they gain an extra 0.0078% chance of winning. For every uber more than the enemy, Trihards gain an extra 0.035% chance of winning. For every drop more than the enemy, their odds DIMINISH by a huge 7.06%. Of course, the values are reversed for every unit less than the enemy.

Also notice that the t-scores of the ubers / drops are fairly small, indicating they're not very relevant (although dropping much moreso that ubering.,) while damage is very significant.

Again, this is probably very wrong, anyone who knows more about this that I do please feel free to chime in.

TL:DR - read the post sucka.
4
#4
SizzlingStats
6 Frags +

a while ago i scraped 20K logs from tf2logs.com (sorry barncow) to see if the granary spawns actually made it more likely for blu to win mid

blu had something like a 0.5% advantage, which would be in the margin of error probably

a while ago i scraped 20K logs from tf2logs.com (sorry barncow) to see if the granary spawns actually made it more likely for blu to win mid

blu had something like a 0.5% advantage, which would be in the margin of error probably
5
#5
3 Frags +

In general this would be cool for casters during pregame. They could talk about how X demo dies a lot but deals TONS of damage, or about how X scout tends to play poorly on X map, etc.
Interesting stuff. This is something I'd bet the whole community would benefit from.

In general this would be cool for casters during pregame. They could talk about how X demo dies a lot but deals TONS of damage, or about how X scout tends to play poorly on X map, etc.
Interesting stuff. This is something I'd bet the whole community would benefit from.
6
#6
0 Frags +

I think something like this could help with something like an invite fantasy tf2 league. I always thought that would be interesting.

I think something like this could help with something like an invite fantasy tf2 league. I always thought that would be interesting.
7
#7
5 Frags +

I SEE NUMBERS WHAT'S HAPPENING SOMEONE EXPLAIN PLEASE?

I SEE NUMBERS WHAT'S HAPPENING SOMEONE EXPLAIN PLEASE?
8
#8
0 Frags +

Depends on how much analysis you're actually willing to do. Some of the stats we have right now don't tell the whole story. For instance, demos and soldiers can pad their stats but not necessarily help their team, or a dropped uber does not tell the whole story.

What you could do is go into a demo and look at specific plays and decide whether or not it was a good play, not just judging by damage or k/d. Here are some examples off the top of my head that are easily attainable but require work:

- key plays (dropping/killing/forcing the med, killing the demo, stopped a backcap, etc)
- backcaps that helped their team win the round
- offclass success rate, probably better for seasonal stats (did the spy/sniper get a key pick? did the heavy/engie/pyro stop a last push? etc.)

If you REALLY wanna nerd out, you can tally up useful damage stats, by subtracting useless damage from total damage to see whos actually putting out good damage (ie. two 300 pockets exchanging rockets at gully choke, baiting your team)

If you wanna do analysis similar to some of the existing sports ones like PER and QB rating, you'd have to come up with a formula that quantifies each and every little thing that a player does and multiply that by a factor, whether negative or positive, then totalling that up into a final score. Again, takes time and effort to come up with a formula that's efficiently analyzes a player's performance.

Depends on how much analysis you're actually willing to do. Some of the stats we have right now don't tell the whole story. For instance, demos and soldiers can pad their stats but not necessarily help their team, or a dropped uber does not tell the whole story.

What you could do is go into a demo and look at specific plays and decide whether or not it was a good play, not just judging by damage or k/d. Here are some examples off the top of my head that are easily attainable but require work:

- key plays (dropping/killing/forcing the med, killing the demo, stopped a backcap, etc)
- backcaps that helped their team win the round
- offclass success rate, probably better for seasonal stats (did the spy/sniper get a key pick? did the heavy/engie/pyro stop a last push? etc.)

If you REALLY wanna nerd out, you can tally up useful damage stats, by subtracting useless damage from total damage to see whos actually putting out good damage (ie. two 300 pockets exchanging rockets at gully choke, baiting your team)

If you wanna do analysis similar to some of the existing sports ones like PER and QB rating, you'd have to come up with a formula that quantifies each and every little thing that a player does and multiply that by a factor, whether negative or positive, then totalling that up into a final score. Again, takes time and effort to come up with a formula that's efficiently analyzes a player's performance.
9
#9
0 Frags +

LANs assisted, LANs won, how long have the player been in the team, overall position in season ranking for frags, overall position in season ranking for damage, hours played, hours on his main played, overall season k/d ratio, Reddit MGE ELO (heh), number of tf2.pug.na pugs played, number of tf2lobby.com lobbies played.

idk, I'm just saying random shit

LANs assisted, LANs won, how long have the player been in the team, overall position in season ranking for frags, overall position in season ranking for damage, hours played, hours on his main played, overall season k/d ratio, Reddit MGE ELO (heh), number of tf2.pug.na pugs played, number of tf2lobby.com lobbies played.

idk, I'm just saying random shit
10
#10
4 Frags +

DOES TRI RUNNING GUNBOATS RESULT IN MORE/LESS FRAGS/DAMMAGE

DOES TRI RUNNING GUNBOATS RESULT IN MORE/LESS FRAGS/DAMMAGE
11
#11
0 Frags +
fsXDRelationships between damage done / kills gotten on classes, just interesting ratios like heals to ubers or damage taken to deaths.

Like, just looking at stats for each player across the whole season is interesting, you could graph out performance levels and see what teams are strong on x maps or could be like how effective is the clockwork factor or something. Stats are usually interesting unless it's something that no one cares about, like the amount of times the esea ads play.

Yes exactly. Good ideas
Gggylgy your stats are spot on.
#5 that's was my motivation
#6 #8 Ain't nobody got time for dat! Unless you have it written in python, in which case import i_have_time.py

[quote=fsXD]Relationships between damage done / kills gotten on classes, just interesting ratios like heals to ubers or damage taken to deaths.

Like, just looking at stats for each player across the whole season is interesting, you could graph out performance levels and see what teams are strong on x maps or could be like how effective is the clockwork factor or something. Stats are usually interesting unless it's something that no one cares about, like the amount of times the esea ads play.[/quote]

Yes exactly. Good ideas
Gggylgy your stats are spot on.
#5 that's was my motivation
#6 #8 Ain't nobody got time for dat! Unless you have it written in python, in which case import i_have_time.py
Please sign in through STEAM to post a comment.