2016-17 Team Win Projection Contest / Discussion

[BasketDork]

Sputtering away from contention over the past month.

[Mike G]

Yeah, but: I've gone from first to nearly last. Bos, Hou, Mia, Was, Uta keep winning. Atl, Det, Ind, LAL, Min, Phx keep on losing.
You are at least better than Vegas in both Avg Err and RMSE -- which is quite something, really.

Currently the average of our 20 guesses would be in 1st place in Avg Err and SMRE; close 2nd in RMSE.
Coincidentally, later entries are tending to rank better than those who were quicker to throw hat into ring.

[BasketDork]

Yeah, I believe I'm a sooner rather than a later.

[Mike G]

It's going down to the wires.

Code: Select all

4/06   RMSE      avg abs err             SMRE
trzu   4.47      amp5   3.79      Crow   3.17
RyRi   4.57      kmed   3.80      kmed   3.26
amp5   4.59      trzu   3.82      amp5   3.29
ncsD   4.68      Crow   3.85      shad   3.33
cali   4.69      RyRi   3.85      yoop   3.35
lnqi   4.71      BaDo   3.93      vegas  3.35
kmed   4.80      lnqi   3.94      sbs.   3.36
AnJo   4.81      sbs.   3.95      trzu   3.38
sbs.   4.83      cali   3.98      RyRi   3.39
ATCt   4.84      ncsD   3.99      BaDo   3.46
BaDo   4.86      yoop   4.00      lnqi   3.55
sndi   4.90      shad   4.01      cali   3.56
yoop   4.98      vegas  4.01      sndi   3.57
vegas  5.00      sndi   4.09      ncsD   3.60
shad   5.07      ATCt   4.11      GK5.   3.61
Crow   5.17      AnJo   4.33      ATCt   3.64
ncsB   5.20      ncsB   4.35      ncsB   3.79
jg34   5.28      GK5.   4.38      nrfo   3.81
Mike   5.44      jg34   4.41      Mike   3.81
taco   5.50      Mike   4.42      jg34   3.94
nrfo   5.50      nrfo   4.48      538    4.03
GK5.   5.55      taco   4.69      AnJo   4.07
538    6.10      538    4.82      taco   4.10

Code: Select all

4/07   RMSE      avg abs err             SMRE
trzu   4.47      kmed   3.73      Crow   3.10
RyRi   4.57      Crow   3.81      kmed   3.12
amp5   4.60      amp5   3.81      vegas  3.21
cali   4.67      BaDo   3.85      BaDo   3.33
ncsD   4.68      trzu   3.85      amp5   3.37
lnqi   4.70      RyRi   3.87      sbs.   3.39
BaDo   4.78      lnqi   3.91      shad   3.42
kmed   4.79      sbs.   3.95      trzu   3.44
AnJo   4.80      ncsD   3.95      RyRi   3.46
sbs.   4.81      vegas  3.95      lnqi   3.47
ATCt   4.84      cali   3.98      yoop   3.51

If we think of these 3 measures as silver-gold-bronze -- average absolute error being the main prize -- we could weight them 2-3-1 and take the avg, to arrive at a single winner.
With that weighting, we get

Code: Select all

3.98   kmed      4.26   shad
3.99   trzu      4.27   sndi
4.00   amp5      4.29   yoop
4.04   RyRi      4.32   ATCt
4.07   BaDo      4.40   AnJo
4.10   lnqi      4.53   ncsB
4.12   ncsD      4.59   jg34
4.13   Crow      4.60   Mike
4.14   sbs.      4.60   GK5.
4.15   cali      4.68   nrfo
4.17   vegas     4.85   taco
.                5.10   538

Weighted 1-1-1, the top 5 are unchanged, kmed with a bigger margin.
My dog isn't in the hunt, and I don't object to seeing 2 or 3 co-winners.

[shadow]

538 might want to rethink switching to a 100% BPM based model from their previous BPM/RPM blend.

[Crow]

Yeah they should. Go back to equal weight or remove BPM altogether. May substitute in PT-PM or Dredge

RMSE likes me less than the other two. Will have to review variance detail after finish.

[Mike G]

Weighted 2-3-1

Code: Select all

wtd sum       RMSE  AvgErr  SMRE
3.98   trzu   4.43   3.85   3.48
4.02   kmed   4.81   3.77   3.20
4.04   amp5   4.61   3.86   3.45
4.07   RyRi   4.56   3.92   3.53
4.08   lnqi   4.64   3.90   3.48
4.09   BaDo   4.79   3.87   3.35
4.12   ncsD   4.70   3.95   3.48
4.14   cali   4.64   3.98   3.61
4.14   Crow   5.05   3.84   3.22
4.14   vegas  4.93   3.93   3.21
4.18   sbs.   4.79   4.00   3.49
4.24   yoop   5.02   4.00   3.43
4.30   sndi   4.95   4.10   3.59
4.33   shad   5.07   4.10   3.54
4.37   AnJo   4.79   4.25   3.88
4.37   ATCt   4.85   4.22   3.87
4.55   Mike   5.30   4.33   3.71
4.56   ncsB   5.20   4.37   3.81
4.57   jg34   5.26   4.36   3.84
4.59   GK5.   5.52   4.33   3.52
4.76   nrfo   5.54   4.54   3.84
4.84   taco   5.47   4.67   4.06
5.11   538    6.12   4.82   3.95

4-11
Same ordering scheme -- 2.3.1 -- but relative to the prediction of 'regressed 2016 pythagorean wins'.

Code: Select all

PyR-xyz       RMSE   AvEr   smre    xyz
1.84    avg   4.56   3.82   3.34    3.99
1.81   trzu   4.44   3.89   3.55    4.01
1.81   kmed   4.87   3.75   3.11    4.02
1.79   cali   4.67   3.86   3.29    4.03
1.76   lnqi   4.78   3.88   3.22    4.07

1.75   BaDo   4.88   3.82   3.22    4.07
1.67   amp5   4.77   3.97   3.48    4.16
1.66   ncsD   4.80   3.98   3.45    4.16
1.63  vegas   5.00   3.96   3.26    4.19
1.61   Crow   5.08   3.93   3.32    4.21

1.60   RyRi   4.69   4.08   3.72    4.22
1.56   sbs.   4.89   4.09   3.54    4.26
1.51   yoop   5.04   4.09   3.57    4.32
1.46   sndi   5.01   4.17   3.64    4.36
1.43   AnJo   4.90   4.25   3.83    4.40

1.34   shad   5.20   4.25   3.75    4.48
1.33   ATCt   5.05   4.32   3.91    4.49
1.30   Mike   5.40   4.27   3.56    4.53
1.27   jg34   5.34   4.33   3.63    4.55
1.11   ncsB   5.30   4.54   4.07    4.72

1.07   GK5.   5.64   4.51   3.75    4.76
0.94   nrfo   5.74   4.65   3.92    4.89
0.86   taco   5.59   4.79   4.21    4.96
0.49    538   6.28   5.05   4.28    5.33
0.00  16PyR   6.62   5.59   4.94    5.82

Note avg does not lead in any category. It's the avg of our 20 active entries, not incl. 538, AnJo, or Vegas.

4-12

Code: Select all

Py-xyz  231   RMSE   AvEr   smre    xyz
1.78   kmed   4.97   3.81   3.12    4.08
1.77   trzu   4.52   3.96   3.60    4.09
1.75   cali   4.76   3.93   3.33    4.11
1.74    avg   4.66   3.96   3.51    4.12
1.74   BaDo   4.97   3.86   3.20    4.12
1.67   lnqi   4.89   4.00   3.34    4.19

1.57   amp5   4.88   4.11   3.64    4.29
1.57   Crow   5.16   4.01   3.42    4.29
1.57   ncsD   4.90   4.11   3.63    4.29
1.54   yoop   5.08   4.09   3.54    4.33
1.53  vegas   5.11   4.10   3.44    4.33
1.51   RyRi   4.80   4.22   3.87    4.35

1.47   sbs.   5.00   4.23   3.70    4.40
1.40   AnJo   4.99   4.32   3.83    4.46
1.36   sndi   5.14   4.31   3.79    4.50
1.29   Mike   5.47   4.30   3.60    4.57
1.25   shad   5.27   4.39   3.93    4.61
1.24   ATCt   5.16   4.46   4.05    4.63

1.21   jg34   5.45   4.43   3.73    4.65
1.02   ncsB   5.38   4.68   4.24    4.84
1.00   GK5.   5.72   4.62   3.86    4.86
0.87   nrfo   5.82   4.75   4.06    4.99
0.77   taco   5.71   4.92   4.37    5.09
0.45    538   6.38   5.13   4.35    5.41

[Mike G]

Final tallies

Code: Select all

4/13   RMSE      avg abs err             SMRE
trzu   4.54      kmed   3.84      BaDo   3.03
cali   4.71      BaDo   3.87      kmed   3.22
RyRi   4.76      cali   3.93      yoop   3.30
ncsD   4.81      yoop   4.00      vegas  3.36
amp5   4.81      trzu   4.00      lnqi   3.42
lnqi   4.86      lnqi   4.00      cali   3.42
AnJo   4.91      vegas  4.06      ncsD   3.49
kmed   4.94      ncsD   4.07      amp5   3.49
BaDo   4.95      amp5   4.07      Crow   3.52
sbs.   4.96      Crow   4.10      Mike   3.57
yoop   5.05      RyRi   4.18      sndi   3.57
vegas  5.09      sbs.   4.22      trzu   3.60
sndi   5.10      AnJo   4.23      AnJo   3.67
ATCt   5.12      sndi   4.27      sbs.   3.70
Crow   5.17      Mike   4.33      jg34   3.75
shad   5.23      shad   4.40      RyRi   3.77
ncsB   5.32      ATCt   4.40      GK5.   3.88
jg34   5.41      jg34   4.43      nrfo   3.90
Mike   5.49      GK5.   4.60      shad   3.96
GK5.   5.65      ncsB   4.60      ATCt   3.99
taco   5.69      nrfo   4.73      538    4.06
nrfo   5.84      taco   4.93      ncsB   4.09
538    6.32      538    5.07      taco   4.38

Note we have 3 different winners. All 3 are 8th or worse in another column.
Weighting them 2-3-1 respectively, we get this composite:

Code: Select all

xyz   3-2-1   RMSE  AvgErr  smre
4.09   BaDo   4.95   3.87   3.03
4.11   kmed   4.94   3.84   3.22
4.11   avg    4.63   3.95   3.53
4.11   cali   4.71   3.93   3.42
4.11   trzu   4.54   4.00   3.60
4.19   lnqi   4.86   4.00   3.42
4.22   ncsD   4.81   4.07   3.49
4.22   amp5   4.81   4.07   3.49
4.23   yoop   5.05   4.00   3.30
4.29   vegas  5.09   4.06   3.36
4.31   RyRi   4.76   4.18   3.77
4.36   Crow   5.17   4.10   3.52
4.36   AnJo   4.91   4.23   3.67
4.38   sbs.   4.96   4.22   3.70
4.43   sndi   5.10   4.27   3.57
4.57   ATCt   5.12   4.40   3.99
4.59   Mike   5.49   4.33   3.57
4.60   shad   5.23   4.40   3.96
4.65   jg34   5.41   4.43   3.75
4.76   ncsB   5.32   4.60   4.09
4.83   GK5.   5.65   4.60   3.88
4.96   nrfo   5.84   4.73   3.90
5.09   taco   5.69   4.93   4.38
5.32   538    6.32   5.07   4.06
5.87  16PyR   6.64   5.63   5.05

Weighting the columns 1-1-1 or 1-2-1, BaDo wins by bigger margins. Weighted 2-2-1, he barely outdoes trzu.
Congrats to the Bearded Geek, BasketDork !

[caliban]

RMSE 2nd place two years in a row is like going 30-11 second half of the season but missing the playoffs.

[Mike G]

On Feb. 3, tarrazu hit 2.87 Avg Err.
The best at the end of the season -- 3.84 (kmedved) -- was 5th worst back then

In the last 3 days, we had 3 different composite leaders.

[permaximum]

SMRE shouldn't have any weight. It doesn't even make sense especially in this "NBA team win projection contest". This is not dart.

Also, using actual wins is a must.

[tarrazu]

SMRE shouldn't have any weight. In doesn't even make sense especially in this "NBA team win projection contest". This is not dart.

Also, using actual wins is a must.

Pretty biased obviously , but either RMSE or RMSE/Avg Error blend seems most reasonable for the idea of projecting win totals.

[bbstats]

RMSE is the standard and should remain unless voted on prior to the contest

*grumbles about prior 2nd place*

[kmedved]

Agreed. I haven't seen a good reason to switch from RMSE yet (though the fact that I grade higher in Average Error seems like a good reason!).

[Mike G]

Unless someone named Ramsey (or Ramsay or Rimsie ...) comes along to endorse RMSE, I don't know how to use it in a sentence.