|
|
Post by 20s on May 2, 2021 11:12:49 GMT -5
Late start this year due to being out of the country. Let’s dive right in.  The league is skewing even more to the offense, not sure at what point we will see this slow down. A large majority of the upgrade points are placed into offensive attributes so that is what drives this.  The Chaps winning the title last year with the 19th ranked defense (and worse than average defense) has ruined the shading of the charts. Now anyone with average or better defense is showing in the shaded area. But the Chaps also had the league’s best offense so clearly a bunch of these teams aren’t actually championship worthy. The Cougars and Warriors are the only two teams in the championship quadrant right now on the ORtg+ and DRtg+ chart. The Bullets are far and away worse than anyone else to this point.  Similar results here, but when we go by league rank instead of how dominant they are, we see the Bucks, Sixers and Pacers enter the quadrant with the Nets right on the line. Again, the Chaps from last year skew this a bit as before they won, the Sixers wouldn’t have been in the quadrant.  The raw O Rtg / D Rtg chart isn’t really valuable at this point due to the offensive rise of the league combined with the Chaps’ defense of last year making the “championship quadrant” very easily attainable for teams in 3005.  With the quadrants becoming questionable/not useful, I decided to try clustering to see how that separated the teams. It naturally separated the league into two clusters. The orange cluster contained all of the previous champions and 11 of the teams from 3005. The blue cluster contains 18 of the teams from 3005 who currently seem less likely to be a contender. This may become a useful tool to move going forward.  |
|
Deleted
Posts: 0
Likes:
Joined: January 1970
|
Post by Deleted on May 2, 2021 11:44:27 GMT -5
What algorithm are you using to cluster? Have you tried more than 2 clusters?
|
|
|
|
Post by 20s on May 2, 2021 12:46:09 GMT -5
What algorithm are you using to cluster? Have you tried more than 2 clusters? It uses k-means clustering. I’m using Tableau and just using their native feature and it uses k-means. Yeah, I did more than 2 as well. I can share some of those results, too. Three clusters just throws the Bullets in a cluster by themselves and keeps everything else the same. |
|
Deleted
Posts: 0
Likes:
Joined: January 1970
|
Post by Deleted on May 2, 2021 12:51:00 GMT -5
What algorithm are you using to cluster? Have you tried more than 2 clusters? It uses k-means clustering. I’m using Tableau and just using their native feature and it uses k-means. Yeah, I did more than 2 as well. I can share some of those results, too. Three clusters just throws the Bullets in a cluster by themselves and keeps everything else the same. Kmeans is great. The result with the three clusters and Bullets makes sense, trying 4 might be more enlightening.
|
|
|
|
Post by 20s on May 2, 2021 12:57:06 GMT -5
It uses k-means clustering. I’m using Tableau and just using their native feature and it uses k-means. Yeah, I did more than 2 as well. I can share some of those results, too. Three clusters just throws the Bullets in a cluster by themselves and keeps everything else the same. Kmeans is great. The result with the three clusters and Bullets makes sense, trying 4 might be more enlightening.
Yeah, it has a feature where you can let it “optimize” the number of clusters instead of specifying and that created two so that is why I posted that. Here is their documentation on how they optimize the number of clusters. Criteria used to determine the optimal number of clusters
Tableau uses the Calinski-Harabasz criterion to assess cluster quality. The Calinski-Harabasz criterion is defined as  where SSB is the overall between-cluster variance, SSW the overall within-cluster variance, k the number of clusters, and N the number of observations. The greater the value of this ratio, the more cohesive the clusters (low within-cluster variance) and the more distinct/separate the individual clusters (high between-cluster variance). Since the Calinski-Harabasz index is not defined for k=1, it cannot be used to detect one-cluster cases. If a user does not specify the number of clusters, Tableau picks the number of clusters corresponding to the first local maximum of the Calinski-Harabasz index. By default, k-means will be run for up to 25 clusters if the first local maximum of the index is not reached for a smaller value of k. You can set a maximum value of 50 clusters. |
|
Deleted
Posts: 0
Likes:
Joined: January 1970
|
Post by Deleted on May 2, 2021 13:29:51 GMT -5
Yeah, it has a feature where you can let it “optimize” the number of clusters instead of specifying and that created two so that is why I posted that. Here is their documentation on how they optimize the number of clusters. Criteria used to determine the optimal number of clusters
...By default, k-means will be run for up to 25 clusters if the first local maximum of the index is not reached for a smaller value of k. You can set a maximum value of 50 clusters.
Nice. I think optimizing clusters is great for a lot of use cases. When you want to be sure each cluster is most similar to each other etc. I think for the purposes here using manual clustering will be better because it should separate the teams into "tiers", and even though some of the tiers might be slightly more unbalanced than others (not maximizing the CH index), it would be great for our purposes in trying to tease out rankings for groups of teams across different situations. Excited to see more of this stuff :D
|
|
|
|
Post by eric on May 2, 2021 14:37:14 GMT -5
yeah really enjoying the clusters approach. i think i used k means making a kinerd location cluster back in the day  |
|
|
|
Post by 20s on May 3, 2021 14:09:39 GMT -5
Sim 7 Update:O Rtg+ & D Rtg+ O Rank & D Rank
 Raw O Rtg & D Rtg
  Two Clusters:
 Seven Clusters (kept going until I thought they were interesting):
 - So the Bullets are in Green all by their lonesome in the corner. Just the way Yawn likes it.
- The Grizzlies and Pipers are the purple teams in their own group for shitty but not as bad as the Bullets.
- The Red Group aka Non Contenders who suck at offense consists of the Spurs, Knicks, Sonics, Kings, Cavs, and Hawks.
- The Blue Group aka Non Contenders who are better at offense includes the Raptors, Hornets, Mavs, Suns, Stags, Heat, and Chaps.
- The Orange group is made up of mostly playoff teams who currently aren't getting it done at the level of previous champions. This group is unlikely to win the championship IMO, but maybe some of the better offenses in the group would have a chance with some RNG fortune. It includes the Bulls, Clippers, Blazers, Magic, Lakers, Timberwolves, Bobcats, and Pacers.
- The best of the best were split into two clusters. Teal and Yellow. These clusters contain the 5 previous champions and 5 current 3005 TMBSL teams.
- Teal: 04 Chaps, 00 Mavs, 01 Grizzlies, and the current 76ers
- Yellow: 02 76ers, 03 76ers and the current Bucks, Nets, Cougars, and Warriors
- The main differentiator between these two groups is the defense. The two 76ers teams to win the titles were the best defenses that have won a championship. The Warriors' defense is in line with those two teams while the other 3 current teams are not quite there but better than the current 76ers.
- The 76ers defense groups them in with the 00 Mavs, 01 Grizzlies, and 04 Chaps. Their offense isn't on the same level as any of those previous champs. They are a team who could win a title, but they likely are going to need to see some improvement from their team post-McCray trade.
- The Warriors, Cougars, Nets, and Bucks all have the profiles of teams who could win a title, but the Bucks and Nets could benefit from improved offense. The Cougars need to recapture their play from the first 40 days of the season and break out of their current funk.
For Statistics Nerds only:
|
|
|
|
Post by 20s on May 3, 2021 14:10:02 GMT -5
New chart coming soon!
|
|
Deleted
Posts: 0
Likes:
Joined: January 1970
|
Post by Deleted on May 3, 2021 14:13:19 GMT -5
this is so fucking cool 20s
|
|
|
|
Post by 20s on May 3, 2021 14:35:23 GMT -5
Scroll left to right to see the full image.
 Scroll left to right to see the full image.
So wanted to look at Net Rating (O Rtg - D Rtg) which is how many points you outscore your opponents by per 100 possessions. Looking at all the teams for every season so far and highlighting the Champs from each year. Couple things stand out:- 3001 76ers and 3004 Cougars are the only teams to lead in Net Rtg and not win the Championship.
- 3005 Bullets are the worst team we have seen in 6.0 by Net Rtg
- Overall, the league seems to be bunching more together the further we move along in the league. Earlier on, you saw some more dominant and weaker teams. This makes sense with an increase in 'good' or talented players in the pool and more equal distribution of talent. This distribution is more what I expected to see in a wheel league.
- In the creation 3000 season, we didn't have any terrible teams. Now in 3006, other than the Bullets, we are kind of back to that point.
- The 03 76ers were the best team by this metric. We have yet to see a team eclipse the +20 mark in Net Rating.
|
|
Deleted
Posts: 0
Likes:
Joined: January 1970
|
Post by Deleted on May 3, 2021 14:37:49 GMT -5
which team is the 04 warriors stuck behind?
|
|
|
|
Post by 20s on May 3, 2021 14:47:54 GMT -5
which team is the 04 warriors stuck behind? The 76ers. I made a chart with only positive Net Rtg teams so we could see stuff more clearly and the 04 Warriors are still hidden :rv:  |
|
Deleted
Posts: 0
Likes:
Joined: January 1970
|
Post by Deleted on May 3, 2021 14:49:44 GMT -5
Love this stuff 20s. The order on these things and the order of my kilbins is very close. It's cool you can look at team performance and draw extremely similar conclusions to looking at the sum of player contribution.
|
|
|
|
Post by Trofie on May 3, 2021 15:06:08 GMT -5
Go bucks!
|
|
|
|
Post by eric on May 3, 2021 20:16:59 GMT -5
So the Bullets are in Green all by their lonesome in the corner. Just the way Yawn likes it. ALRIGHT WAY TO GO DONNIE Yawn of the Dead |
|
|
|
Post by eric on May 3, 2021 20:18:44 GMT -5
loving the graphics you're putting out, but i'm not sure the league bunching more together hypothesis holds up. if we drop the lowest and highest teams and look at the other 27, the range looks consistent to me over time (except for 3003 for whatever reason)
|
|
|
|
Post by 20s on May 3, 2021 21:00:39 GMT -5
loving the graphics you're putting out, but i'm not sure the league bunching more together hypothesis holds up. if we drop the lowest and highest teams and look at the other 27, the range looks consistent to me over time (except for 3003 for whatever reason) Yeah, I didn’t spend a lot of time on the “analysis” portion of the new chart. That was just what I noticed at first glance, perhaps it is incorrect. |
|
|
|
Post by Yawn of the Dead on May 4, 2021 13:38:01 GMT -5
FINALLY
|
|
|
|
Post by 20s on May 4, 2021 16:22:45 GMT -5
|
|
|
|
Post by 20s on May 7, 2021 17:02:00 GMT -5
Sim 11 Update:
O Rtg+ and D Rtg+
O Rank and D Rank
Raw O Rtg and D Rtg
Three Clusters (Shit, Okay, Contenders)
More to come...
|
|
|
|
Post by skrouse on May 7, 2021 17:13:55 GMT -5
|
|
|
|
Post by TinyTimPig on May 7, 2021 18:51:53 GMT -5
Let’s maybe chill out with the cluster names
|
|
bankz
New Member
GM of the Year: 3001
Posts: 2,347
Likes: 490
Joined: January 2021
|
Post by bankz on May 7, 2021 18:58:19 GMT -5
I like the names. fits the teams listed
|
|
|
|
Post by 20s on May 11, 2021 14:49:03 GMT -5
|
|
|
|
Post by Trofie on May 11, 2021 14:50:16 GMT -5
120k
|
|
|
|
Post by TinyTimPig on May 11, 2021 14:52:11 GMT -5
100k
|
|
bankz
New Member
GM of the Year: 3001
Posts: 2,347
Likes: 490
Joined: January 2021
|
Post by bankz on May 11, 2021 15:01:43 GMT -5
100K
|
|
duc15
New Member
Posts: 171
Likes: 45
Joined: January 2021
|
Post by duc15 on May 11, 2021 15:09:28 GMT -5
100k
|
|
IanBoyd
New Member
Posts: 792
Likes: 252
Joined: December 2020
|
Post by IanBoyd on May 11, 2021 15:37:09 GMT -5
150k
|
|
