Jump to content
bjshnog

⟪WN8⟫ Development / Resources

Recommended Posts

Noobmeter is still using the old formula, haven´t had a reply yet since I sent Mr. NM the new formula with frag cap and winrate term.

Link to post
Share on other sites

Checked tank table for last time:

-kv13: i know tank had major rebalances, but i am surprised its stats are that much lower than rest of tier 7 meds? Is it because its unpopular or because it sucks? Played it long ago, really no idea.

-tier 6 arty is suspicious. Folterknecht?

-i have some concerns over t18 not standing out from other tds. Played it yesterday to collect 50 games, thing is ridiculously OP.

- i can see 263 as good padding tank, but those who grind useless turds before it deserve it

 

 

Nice to see most of mine earlier objections corrected themselves with filtering out baddies hehe.

Link to post
Share on other sites
...

-tier 6 arty is suspicious. Folterknecht?

...

 

1600+dpb - if you grind these things up even with 100% crew that seems now impossible while you are top tier in many matches. I would bring that down a bit. If you play them just for fun in platoons with t7-8 tanks doable.

 

Edit:

 

Even the KV1s only has ~1400dpb for 2400 rating ... . I think thats easier to do than 1600+dpb in nowdays t6 artis. Throw in some history malus and put them to around 1550.

Link to post
Share on other sites

Because i doubt those numbers were "that easy" even when they were tier 5, let alone now. I guess we can just  gradually lower arty stats with time as amount of new battles (post nerf) increases.

Link to post
Share on other sites

Because i doubt those numbers were "that easy" even when they were tier 5, let alone now. I guess we can just  gradually lower arty stats with time as amount of new battles (post nerf) increases.

 

Hopefully, as time passes, the database we're sourcing from will be settled, and no manual processes will be needed. 

Link to post
Share on other sites

Because i doubt those numbers were "that easy" even when they were tier 5, let alone now. I guess we can just  gradually lower arty stats with time as amount of new battles (post nerf) increases.

 

1500 dpb with t5 arti was very good at least and 1600+ unicum territory, no discussion there.

Link to post
Share on other sites

Crab - hit me up if you need eyes over your post.

 

When I'm looking at the histogram, I'm seeing about the following:

 

<400

400-600

600-800

800-1000

1000-1200

1200-1400

1400-1700

1700-2300

2300-2900

>2900

 

 

 

 

Link to post
Share on other sites

Crab - hit me up if you need eyes over your post.

 

When I'm looking at the histogram, I'm seeing about the following:

 

<400

400-600

600-800

800-1000

1000-1200

1200-1400

1400-1700

1700-2300

2300-2900

>2900

 

Affirmative.

Link to post
Share on other sites

I changed the values in my post a little, so updated your quote too. Heres that scale in semi-graph form.

 

1Ji9gIq.png

 

What do you guys think about adjusting the scale so that it's more 60 day friendly, or using a separate scale for 60 day (maybe just all values + 200 for 60d). 60d stats are inherently much higher than overall, and thats broken the scales for WN7 on the colour signatures.

Link to post
Share on other sites

I changed the values in my post a little, so updated your quote too. Heres that scale in semi-graph form.

1Ji9gIq.png

What do you guys think about adjusting the scale so that it's more 60 day friendly, or using a separate scale for 60 day (maybe just all values + 200 for 60d). 60d stats are inherently much higher than overall, and thats broken the scales for WN7 on the colour signatures.

That graph looks pretty and real easy to understand. Gj

Link to post
Share on other sites

Well, I think the winrate part is actually accounting more for the 10% of rWINc that Eureqa could not explain using rFRAGc rDAMAGEc, rSPOTc and rDEFc. After including it, the curve is a little bit less steep, which is good I think. When not including the winrate term, we would be overrating players on the top, since their stats are very high. Don´t know if that was a good explanation...

 

look back to when I posted the formula eureqa gave me. It had a constant value of around 300 being added in to the equation. That is a constant value that had to be added to the formula using all the stats, to really optimize the correlation with rWINc. That meant that there is a portion of rWINc that Eureqa can[t explain with the stats we were giving it. This, along with Twistoon´s reasoning of platoon vs solo stats, convinced me that putting wins back into the equation was a good idea.

 

 

Also, making winrate term depend on avg tier is not necessary, since it already takes into account that a player of equal skill will win more in the lower tiers.

 

 

Shnog, Marius emailed me too, I am sorry but I do not have the time to dedicate to WN8D. After WN8 is ready, though, we can make a new thread about WN8D and I can help you a bit to see if anyone is up to the challenge.

 

But still, platoon players will cause heaps of distortion. It should only represent what stats people missed out on because of their good platoonmates taking them.

 

More specifically, it should focus on the difference between expected rWINc for that person's stats and their actual rWINc. If both of those values are equal (rWINc = 1), then the win rate term should be equal to zero. If their actual win rate is higher, then they should get a bonus. IMPORTANT: The bonus/malus must vary on tier. At low average tiers (1-4), players can probably miss out on about 30% of the stats if they platoon with good players (rWINc = 2) while at high tiers, it's probably closer to about 10%. If they win LESS, it is probably because they are platooning with bad players, which allows them to deal more damage due to the incompetence on the team at higher tiers. At low tiers, however (1-4), platooning with bad players (resulting in rWINc = 0 or WR = 42%) would probably make next to zero difference to their stats. At high tiers (8-10), they would probably be able to deal about 5% to 15% more damage and kills, etc.

 

HOWEVER, I don't think platooning at low tiers should really be rewarded (30% vs 10%), because right now, the rating is already biased towards high tiers by the fact that they simply have higher expected damage.

 

Basically, making it account for 10% overall will cause too many problems, the main one being platoons being massively overestimated.

 

EDIT: The most important part of this is to find the best curve (actually surface, because it is a function of two variables) which represents (and negates) the change expected in the other stats due to variation in average tier and difference between account rWINc and expected rWINc for the player's stats. Use the original Eureqa solution for rWINc and then compare that to actual rWINc to find the difference.

Link to post
Share on other sites

Winrate term is small enough that it doesn´t really need any more fidgeting I think.

 

For the moment, I really can´t find any better way to apply the winrate term from the data. Open to more ideas though. I tried what you did, and the difference is normally between 30 and 50 points... however, there are some outliers in the low-mid WN8 player range, so I don´t think it is a very good idea. I would rather just leave WR term as it is for the moment.

Link to post
Share on other sites

I did an estimation of the categories for large DBs, since this is a database made up of 10k+ game players, it looks like this (maintaining the same categories as wotlabs):

 

<300 Bot
300-599 BAD
600-899 BELOW AVERAGE
900-1199 ABOVE AVERAGE
1200-1499 GOOD 
1500-1749 VERY GOOD
1750-2299 GREAT
2300-2899 UNICORN
2900-4000 SUPERUNICORN

4000+ RAINBOW ULTRAUNICORNMEGATRON

Link to post
Share on other sites

I don't think it's good enough just to use raw rWINc though...

 

Perhaps the reason for the outliers is simply that those players platoon or do TCs or CW A LOT.

Link to post
Share on other sites

Probably. However, if the winrate term will be proportional to the difference between rWINc and WN8, and we add that in, we will end up having a WN8 rating which is 100% identical to rWINc. We should just use rWINc as a metric. :)

 

I did rWINc*7000 - WN8

 

Players who always play alone should get negative numbers, Zakaladas gets -350, Redparadize gets 5500, and many G players get from 1500 to 6000. This probably measures how much people platoon, but it also measures spotted damage, etc. Can we tell apart the two? No we can´t. However, doesn´t Zakaladas have good intangibles? I am sure he does. His map awareness, tactical area denial, etc. are tip top, however, he gets negative value, one of the lowest actually, at -300.

 

I really can´t read anything from this that would allow a better, more accurate WR term. 

Link to post
Share on other sites

Probably. However, if the winrate term will be proportional to the difference between rWINc and WN8, and we add that in, we will end up having a WN8 rating which is 100% identical to rWINc. We should just use rWINc as a metric. :)

 

I did rWINc*7000 - WN8

 

Players who always play alone should get negative numbers, Zakaladas gets -350, Redparadize gets 5500, and many G players get from 1500 to 6000. This probably measures how much people platoon, but it also measures spotted damage, etc. Can we tell apart the two? No we can´t. However, doesn´t Zakaladas have good intangibles? I am sure he does. His map awareness, tactical area denial, etc. are tip top, however, he gets negative value, one of the lowest actually, at -300.

 

I really can´t read anything from this that would allow a better, more accurate WR term. 

 

Try this: WN8 (with WR) = [0.9 + 0.1 * (rWINc / eWIN)] * WN8, where eWIN = expected rWINc for the players stats (the original Eureqa solution). This reduces total noobs getting comparatively large bonuses.

 

If that doesn't look right, just adjust the 0.9 and 0.1 to something that fits better (ie. 0.93 and 0.07).

 

EDIT: Turns out 0.9 and 0.1 only makes an absolutely insignificant difference. Maybe 0.8 and 0.2? Then again, maybe it should be insignificant for most players only noticeable with outliers.

 

EDIT 2: Yeah, it turns out this form for the win rate results in exactly the same thing as just having the currently-in rWINc term that you posted. Maybe this instead: WN8 (with WR) = [0.9 + 0.1 * sqrt(rWINc / eWIN)] * WN8. This way, you won't need to cap it, it will be smoother and won't massively reward noobs for platooning like yours would. :)

Link to post
Share on other sites

Also, if we are to include rWINc in the formula, the WINS values in the table really need to be tweaked in a way in which we are all certain will return accurate results. I mean, 46.14% in the JPE100? Nope.

Link to post
Share on other sites

WG in patch 0.8.9 will add 7/42 mode to the game. If the statistics will be not separate, it'll be the end of all our efforts :(

 

UP. The developers answered that statistics will be separate :-)

Link to post
Share on other sites

WG in patch 0.8.9 will add 7/42 mode to the game. If the statistics will be not separate, it'll be the end of all our efforts :(

 

I guess it will be tank company change and since very few play TC anyway i doubt it will affect database that much. Expecting equal influence to tier 10 players lowering their damage in clan wars.

 

Edit: replied before your edit =D

 

Link to post
Share on other sites

The developers answered that statistics will be separate :-)

 

Praised be SerB.

Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.

×
×
  • Create New...