Page 11 of 13
Re: 2017-18 team win projection contest
Posted: Tue Feb 27, 2018 4:59 pm
by Crow
Jacob G., is there some blend of PIPM and RPM that beats everything? How does 50/50 blend do? Is there a notable difference how each does separately for good / bad teams, major / minor players?
Re: 2017-18 team win projection contest
Posted: Tue Feb 27, 2018 5:05 pm
by Crow
Separation of top & bottom is 1: that is small. How much is random vs. sustainable precision difference? Maybe there ought to be a multi-year projection contest in future for entries that can do that.
Or less complicated, use existing projections to run the playoff brackets after they are determined.
Re: 2017-18 team win projection contest
Posted: Thu Mar 01, 2018 2:38 pm
by Mike G
A new [co] leader
Code: Select all
avg abs err RMSE avg abs err RMSE
cali 4.52 5.71 kmed 4.87 5.97
sbs. 4.54 5.57 Nath 4.93 6.04
RyRi 4.58 5.54 emin 5.02 5.90
538_ 4.61 5.82 GK5. 5.02 6.47
lnqi 4.61 5.65 ATCt 5.10 6.50
shad 4.69 5.96 Crow 5.13 6.22
sndi 4.75 5.52 Josh 5.15 6.16
vegas 4.76 5.85 knar 5.31 6.51
AJBk 4.84 5.95 Mike 5.42 6.42
gold 4.85 6.06 ncs. 5.59 6.88
Top 5's for various exponents:
Code: Select all
exp = 0.15 exp = 0.70 exp = 1.30
538_ 2.74 cali 4.04 sbs. 4.87
GK5. 2.81 538_ 4.14 RyRi 4.90
cali 2.92 sbs. 4.18 cali 4.94
AJBk 2.95 lnqi 4.20 lnqi 4.96
RyRi 3.02 RyRi 4.21 sndi 5.00
exp = 1.77 exp = 5.00 exp = 10.0
RyRi 5.34 sndi 7.12 emin 8.58
sbs. 5.35 RyRi 7.35 sndi 8.61
sndi 5.36 lnqi 7.38 lnqi 8.68
lnqi 5.44 emin 7.43 RyRi 8.71
cali 5.48 sbs. 7.56 AJBk 8.94
Re: 2017-18 team win projection contest
Posted: Sat Mar 03, 2018 12:50 pm
by Mike G
That didn't last long
Code: Select all
avg abs err RMSE avg abs err RMSE
RyRi 4.69 5.78 Nath 5.08 6.24
sbs. 4.75 5.83 GK5. 5.08 6.69
cali 4.77 5.96 gold 5.10 6.28
538_ 4.78 5.99 emin 5.10 6.07
lnqi 4.82 5.92 Crow 5.28 6.44
shad 4.84 6.20 Josh 5.28 6.36
sndi 4.89 5.73 ATCt 5.32 6.76
vegas 4.91 6.09 knar 5.53 6.78
AJBk 4.96 6.15 Mike 5.59 6.66
kmed 4.97 6.19 ncs. 5.72 7.11
The 4.69 leading number is highest/worst since Dec. 24, also the last day sndesai was on top.
RyRi then rode the top spot down to the season's low of 3.79 on Jan. 4. He and sbs then were 1-2 thru Feb.
Re: 2017-18 team win projection contest
Posted: Sun Mar 04, 2018 9:25 pm
by Crow
Best avg. absolute error showing worse right now than best from end of last season but still time to tighten up. Less than 1 difference. Not sure I really care that much about the last win error on avg. Looks like pretty good to good to me.
Re: 2017-18 team win projection contest
Posted: Thu Mar 08, 2018 11:00 am
by Mike G
Code: Select all
avg abs err RMSE avg abs err RMSE
538_ 5.05 6.34 GK5. 5.33 7.04
RyRi 5.05 6.18 emin 5.33 6.38
sbs. 5.08 6.24 Nath 5.40 6.57
cali 5.10 6.31 Josh 5.57 6.74
sndi 5.17 6.08 Crow 5.58 6.79
kmed 5.18 6.48 gold 5.62 6.78
lnqi 5.21 6.33 ATCt 5.62 7.06
shad 5.24 6.64 Mike 5.71 6.89
vegas 5.26 6.52 knar 5.78 7.05
AJBk 5.26 6.52 ncs. 6.07 7.38
Last time the leader was over 5 was early Dec. Back then, last place was over 7.
Leaders tend to be those who have avoided major misses. Risk taking extremists get pummeled.
Code: Select all
. Best Worst
538_ Atl,Was,GSW,LAL Orl,Tor
RyRi Chi,NYK
sbs. Det,Mia
cali Det,Was
sndi Cha,NYK,SAS Chi
kmed Phx,Por,Sac
. Cha,Det,NYK,Was,NOP
lnqi Atl,Sac NOP
shad Mia,NYK Hou,Mem
AJBk Ind,Mia,Mil,Min,OKC,Sac
. Atl,LAC
GK5. Chi,Cle,Mia,Mil,GSW,Uta
. Phl,Hou,Phx,SAS
emin Cha,Orl,Tor Den,Min,Por
Nath Cha,Phl Ind,Was
Josh Den,Uta Bos,Por,Sac
Crow Atl,Dal,Phx Was
gold Atl,Cle Ind,Mia,NOP,OKC
ATCt .. Cha
Mike Mia,Hou,LAC,Min,NOP,SAS
. Brk,Det,Dal,GSW,Uta
knar Brk,Mem,Phx Det
ncs. .. Bos,Cle,GSW,LAC,LAL
Ties are also listed (rounded to integer).
UPDATE Mar. 13
Code: Select all
avg abs err RMSE avg abs err RMSE
538_ 5.20 6.69 GK5. 5.60 7.38
sndi 5.31 6.37 Nath 5.64 6.88
cali 5.32 6.65 AJBk 5.70 6.94
sbs. 5.33 6.63 ATCt 5.79 7.38
RyRi 5.33 6.55 Crow 5.86 7.14
kmed 5.34 6.79 gold 5.88 7.10
lnqi 5.38 6.67 Josh 5.94 7.10
emin 5.41 6.61 knar 5.99 7.42
shad 5.50 7.03 Mike 6.00 7.27
vegas 5.51 6.90 ncs. 6.30 7.76
Re: 2017-18 team win projection contest
Posted: Wed Mar 14, 2018 11:05 am
by Mike G
Raptors and Rockets each with 1/3 chance of winning it all, according to b-r.com
Showing prob of reaching the Finals; winning it; and chance of winning IF they get there:
Code: Select all
C tm reach win W/R
e Tor .636 .327 .514
w Hou .493 .327 .663
w GSW .352 .221 .628
e Bos .179 .052 .29
e Phl .068 .014 .21
w Uta .032 .012 .38
w OKC .027 .009 .33
w Min .036 .008 .22
w Por .026 .007 .27
e Ind .039 .006 .15
e Was .031 .005 .16
e Cle .020 .003 .15
https://www.basketball-reference.com/fr ... _prob.html
Re: 2017-18 team win projection contest
Posted: Wed Mar 14, 2018 1:21 pm
by nbacouchside
Turns out that screwing up your aging curve has a pretty big effect!
Mike G wrote:Code: Select all
avg abs err RMSE avg abs err RMSE
538_ 5.05 6.34 GK5. 5.33 7.04
RyRi 5.05 6.18 emin 5.33 6.38
sbs. 5.08 6.24 Nath 5.40 6.57
cali 5.10 6.31 Josh 5.57 6.74
sndi 5.17 6.08 Crow 5.58 6.79
kmed 5.18 6.48 gold 5.62 6.78
lnqi 5.21 6.33 ATCt 5.62 7.06
shad 5.24 6.64 Mike 5.71 6.89
vegas 5.26 6.52 knar 5.78 7.05
AJBk 5.26 6.52 ncs. 6.07 7.38
Last time the leader was over 5 was early Dec. Back then, last place was over 7.
Leaders tend to be those who have avoided major misses. Risk taking extremists get pummeled.
Code: Select all
. Best Worst
538_ Atl,Was,GSW,LAL Orl,Tor
RyRi Chi,NYK
sbs. Det,Mia
cali Det,Was
sndi Cha,NYK,SAS Chi
kmed Phx,Por,Sac
. Cha,Det,NYK,Was,NOP
lnqi Atl,Sac NOP
shad Mia,NYK Hou,Mem
AJBk Ind,Mia,Mil,Min,OKC,Sac
. Atl,LAC
GK5. Chi,Cle,Mia,Mil,GSW,Uta
. Phl,Hou,Phx,SAS
emin Cha,Orl,Tor Den,Min,Por
Nath Cha,Phl Ind,Was
Josh Den,Uta Bos,Por,Sac
Crow Atl,Dal,Phx Was
gold Atl,Cle Ind,Mia,NOP,OKC
ATCt .. Cha
Mike Mia,Hou,LAC,Min,NOP,SAS
. Brk,Det,Dal,GSW,Uta
knar Brk,Mem,Phx Det
ncs. .. Bos,Cle,GSW,LAC,LAL
Ties are also listed (rounded to integer).
UPDATE Mar. 13
Code: Select all
avg abs err RMSE avg abs err RMSE
538_ 5.20 6.69 GK5. 5.60 7.38
sndi 5.31 6.37 Nath 5.64 6.88
cali 5.32 6.65 AJBk 5.70 6.94
sbs. 5.33 6.63 ATCt 5.79 7.38
RyRi 5.33 6.55 Crow 5.86 7.14
kmed 5.34 6.79 gold 5.88 7.10
lnqi 5.38 6.67 Josh 5.94 7.10
emin 5.41 6.61 knar 5.99 7.42
shad 5.50 7.03 Mike 6.00 7.27
vegas 5.51 6.90 ncs. 6.30 7.76
Re: 2017-18 team win projection contest
Posted: Wed Mar 14, 2018 1:55 pm
by eminence
Blazing up the rankings, woohoo, wonder if it's mostly my Jazz getting closer to expactations or what.
Re: 2017-18 team win projection contest
Posted: Thu Mar 15, 2018 12:06 pm
by Mike G
That has certainly helped. Other improving teams that favor your positions: GSW, LAC
Also teams getting worse: Atl, Brk, Det, NYK, Dal, Phx
Despite having 11 of the most extreme guesses, you have only 3 of the worst for a team.
Your Tor prediction of 56 is the biggest outlier on the board, fully 8 wins above the avg and 5 higher than anyone else. It's one of 4 that you have a best guess.
At any exponent >7, you would be at #1; currently 3rd in RMSE (e=2).
Re: 2017-18 team win projection contest
Posted: Thu Mar 15, 2018 1:19 pm
by eminence
Mike G wrote:That has certainly helped. Other improving teams that favor your positions: GSW, LAC
Also teams getting worse: Atl, Brk, Det, NYK, Dal, Phx
Despite having 11 of the most extreme guesses, you have only 3 of the worst for a team.
Your Tor prediction of 56 is the biggest outlier on the board, fully 8 wins above the avg and 5 higher than anyone else. It's one of 4 that you have a best guess.
At any exponent >7, you would be at #1; currently 3rd in RMSE (e=2).
Thanks for breaking it down Mike!
Only payed attention to the teams getting better, it's just not much fun to follow those lower teams at this point in the season (for me).
And thanks Toronto!
Re: 2017-18 team win projection contest
Posted: Fri Mar 16, 2018 1:04 pm
by GK5
Mike G wrote:Code: Select all
avg abs err RMSE avg abs err RMSE
538_ 5.05 6.34 GK5. 5.33 7.04
RyRi 5.05 6.18 emin 5.33 6.38
sbs. 5.08 6.24 Nath 5.40 6.57
cali 5.10 6.31 Josh 5.57 6.74
sndi 5.17 6.08 Crow 5.58 6.79
kmed 5.18 6.48 gold 5.62 6.78
lnqi 5.21 6.33 ATCt 5.62 7.06
shad 5.24 6.64 Mike 5.71 6.89
vegas 5.26 6.52 knar 5.78 7.05
AJBk 5.26 6.52 ncs. 6.07 7.38
Last time the leader was over 5 was early Dec. Back then, last place was over 7.
Leaders tend to be those who have avoided major misses. Risk taking extremists get pummeled.
Code: Select all
. Best Worst
538_ Atl,Was,GSW,LAL Orl,Tor
RyRi Chi,NYK
sbs. Det,Mia
cali Det,Was
sndi Cha,NYK,SAS Chi
kmed Phx,Por,Sac
. Cha,Det,NYK,Was,NOP
lnqi Atl,Sac NOP
shad Mia,NYK Hou,Mem
AJBk Ind,Mia,Mil,Min,OKC,Sac
. Atl,LAC
GK5. Chi,Cle,Mia,Mil,GSW,Uta
. Phl,Hou,Phx,SAS
emin Cha,Orl,Tor Den,Min,Por
Nath Cha,Phl Ind,Was
Josh Den,Uta Bos,Por,Sac
Crow Atl,Dal,Phx Was
gold Atl,Cle Ind,Mia,NOP,OKC
ATCt .. Cha
Mike Mia,Hou,LAC,Min,NOP,SAS
. Brk,Det,Dal,GSW,Uta
knar Brk,Mem,Phx Det
ncs. .. Bos,Cle,GSW,LAC,LAL
Ties are also listed (rounded to integer).
UPDATE Mar. 13
Code: Select all
avg abs err RMSE avg abs err RMSE
538_ 5.20 6.69 GK5. 5.60 7.38
sndi 5.31 6.37 Nath 5.64 6.88
cali 5.32 6.65 AJBk 5.70 6.94
sbs. 5.33 6.63 ATCt 5.79 7.38
RyRi 5.33 6.55 Crow 5.86 7.14
kmed 5.34 6.79 gold 5.88 7.10
lnqi 5.38 6.67 Josh 5.94 7.10
emin 5.41 6.61 knar 5.99 7.42
shad 5.50 7.03 Mike 6.00 7.27
vegas 5.51 6.90 ncs. 6.30 7.76
Were my results as extreme as Eminence's? It seems like I had one of the most best/worst guesses on the board. My avg abs err is mid tier, but I'm getting killed in RMSE.
Re: 2017-18 team win projection contest
Posted: Fri Mar 16, 2018 11:38 pm
by Mike G
At the 4th power, you're in last place. At the 1/2 power, you're in first!
Here are our avg errors, your errors, and yours relative to avg.
On the left are teams that you guessed better than our avg -- 17 of 30 teams.
On the right are the 13 where you're currently looking worse than avg.
Code: Select all
tm avg GK5 diff tm avg GK5 diff
GSW 5 1 -4.5 Phl 11 17 6.1
Okl 4 0 -3.9 Phx 5 10 5.3
Cle 8 5 -3.3 SAS 7 12 4.7
Mil 3 0 -3.0 Dal 4 8 3.5
Chi 3 1 -2.6 Hou 6 9 3.2
LAL 10 7 -2.4 Den 3 6 3.0
Uta 4 2 -2.4 Atl 3 5 1.4
Sac 2 1 -1.9 Por 8 9 1.2
Det 2 0 -1.8 LAC 4 5 0.7
NOP 4 3 -1.8 Mem 14 14 0.4
Brk 2 1 -1.6 Tor 13 13 0.3
Mia 2 0 -1.4 Cha 8 8 0.2
Was 2 1 -1.0 Orl 7 7 0.1
Ind 14 13 -0.7
NYK 2 1 -0.4
Bos 7 6 -0.2
Min 2 2 -0.2
totals -33.1 30.1
Not only are you 'winning' 57% of them, you're 13-8 in absolute differences >1.0; AND your total of Both columns is -3.0, which is .10 per team better than avg.
BUT your 3 biggest are all on the wrong side of the ledger. Squaring all, those 3 -- Phl, Phx, SA -- entirely wipe out the 23 'successes'.
Average is a tough standard here. Vegas is avg.
Re: 2017-18 team win projection contest
Posted: Sat Mar 17, 2018 11:32 am
by Mike G
Shuffling in the tight upper echelon and in the upper bottoms. And the whole spread is now just .95 top to bottom.
Code: Select all
avg abs err RMSE avg abs err RMSE
538_ 5.35 6.83 Nath 5.67 6.97
emin 5.36 6.66 GK5. 5.67 7.49
sndi 5.39 6.52 AJBk 5.91 7.15
RyRi 5.45 6.71 ATCt 5.91 7.50
cali 5.45 6.78 Mike 5.96 7.38
sbs. 5.45 6.79 Josh 6.00 7.17
kmed 5.47 6.92 Crow 6.06 7.32
vegas 5.56 7.02 knar 6.12 7.54
lnqi 5.60 6.81 gold 6.13 7.31
shad 5.66 7.18 ncs. 6.30 7.87
Leaders within range of exponents:
Code: Select all
exp/root lead
.01 to .55 GK5
.56 - 1.04 538
1.05 - 1.13 emin
1.14 - 5.6 sndi
5.7 + emin
Re: 2017-18 team win projection contest
Posted: Sat Mar 17, 2018 11:48 am
by eminence
Woohoo! If the season could end today that'd make me feel pretty good
