APBRmetrics

Posted: **Wed Nov 30, 2016 4:16 pm**

Not sure I'd say their defensive issues were overblown considering they currently rank 29th in AdjDRtg on bbr, but it is still a little early I suppose. Although I wouldn't bet on a substantial improvement given their personnel and coaching style.

Posted: **Fri Dec 02, 2016 1:20 pm**

A few days ago, 3 guesses were less than 4.00 avg error, with RyRi at 3.93
Now we've all drifted away, but for the regressed 16Pyth -- which is creeping close to the middle of the field.

Code: Select all

RyRi   4.37   .47      ncsB   4.78   .06
Mike   4.40   .43      jg34   4.82   .01
sndi   4.44   .39      16py   4.84   .00
amp5   4.44   .39      ncsD   4.89  -.05
sbs.   4.47   .37      BaDo   4.92  -.08
lnqi   4.61   .23      GK5.   4.92  -.08
trzu   4.66   .17      ATCt   4.97  -.13
kmed   4.68   .16      cali   5.02  -.18
taco   4.70   .13      Crow   5.07  -.23
shad   4.71   .13      yoop   5.15  -.31
vegas  4.74   .10      nrfo   5.30  -.46

In squared errors, 16py ranks 8th, just ahead of vegas.

Posted: **Sat Dec 03, 2016 11:57 am**

Can't say which is more startling -- our sudden reversal of fortunes as a group; or the disparity in ranking with different exponents applied.

Code: Select all

average absolute errors
RyRi   4.55   .61     vegas   5.01    .15
sbs.   4.66   .50      taco   5.09    .07
amp5   4.66   .50      GK5.   5.16    .00
sndi   4.73   .43      16py   5.16    .00
shad   4.74   .42      ATCt   5.18   -.02
Mike   4.75   .41      jg34   5.19   -.03
ncsB   4.81   .35      cali   5.22   -.06
kmed   4.93   .23      ncsD   5.25   -.09
trzu   4.93   .23      Crow   5.29   -.13
lnqi   4.97   .19      nrfo   5.45   -.29
BaDo   5.00   .16      yoop   5.53   -.37


square root of avg of errors-squared RMSE
Mike   5.54   .60      16py   6.14    .00
RyRi   5.73   .41      shad   6.17   -.03
sbs.   5.78   .36      trzu   6.17   -.03
sndi   5.84   .31      ncsD   6.20   -.06
amp5   5.88   .26      GK5.   6.36   -.22
lnqi   5.96   .19      Crow   6.42   -.28
vegas  5.97   .17      cali   6.42   -.28
jg34   6.05   .09      ATCt   6.43   -.29
ncsB   6.07   .07      BaDo   6.43   -.29
taco   6.09   .05      yoop   6.66   -.52
kmed   6.10   .04      nrfo   6.83   -.68


square of avg of sqrt-errors (SMRE)
RyRi   3.80   .70     vegas   4.45    .05
shad   3.85   .64      ATCt   4.45    .04
sbs.   3.86   .63      GK5.   4.49    .00
amp5   3.90   .59      16py   4.49    .00
ncsB   4.05   .45      cali   4.51   -.01
sndi   4.07   .43      nrfo   4.56   -.06
BaDo   4.09   .40      taco   4.56   -.06
trzu   4.22   .27      jg34   4.56   -.07
Mike   4.22   .27      Crow   4.64   -.15
kmed   4.32   .17      ncsD   4.71   -.22
lnqi   4.35   .14      yoop   4.82   -.33

Posted: **Sat Dec 03, 2016 4:49 pm**

Why am I so low in RMSE ? Not a criticism, Im just curious.

Posted: **Sat Dec 03, 2016 7:15 pm**

Two words: Lakers and Bucks.
And Mavs.
You had the lowest guesses on both. Lakers are headed for 37 wins vs your guess of 21.
Bucks look like a 42-win team vs your call of 28.
Your absolute 'errors' are 16.2 and 13.7, and the squares are 262 and 188, respectively.

Then you were a couple wins higher than most on Dallas -- 40 vs their projection of 27, an error squared of 174.
These 3 misses account for 50% of your squared error total of 1240.
It's quite possible all 3 teams will drift toward their pre-season expectations.

I don't really understand how it adds any understanding to exaggerate these errors by squaring them.
Between the two of us, you have 6 of the 7 best guesses and 3 of the 7 worst. And you rank better in SMRE

You show zero (rounded) squared errors for Cha, Hou, Mia, Phx, Sac, and Uta. That's <.707 abs err, and 6 such exact hits seems to lead the field.

Posted: **Sat Dec 03, 2016 8:20 pm**

With the season at about 23%, for the record, here are the prognosticators' best and worst (relative to the field) guesses:

Code: Select all

.      closest on           furthest on
RyRi  Atl Sac
shad  Atl Mia Min Phl      Tor Sac   
sbs.  Atl OKC Sac  
amp5  Atl Den OKC Uta
ncsB  Brk NYK Phl          LAL Orl Por Uta

sndi  SAS                  Min
BaDo  Cha Hou Mia Phx Sac  Brk LAL Mil 
trzu  Det Mil Phl SAS      Bos Por Tor
Mike  Bos Chi LAL NOP      Hou Ind SAS
kmed  Hou Was              Min

lnqi  Phx Sac
ATCt  Ind Mem
GK5.  Chi Cle Tor Was      GSW NYK Phl
cali  OKC Phl Uta          Sac SAS
nrfo  Atl Orl Phx          Cle Dal Den Det SAS

taco  GSW Ind              Cle Orl
jg34  Sac                  Tor
Crow  LAC Orl Uta          Atl Mem Phx Tor Was
ncsD  Uta                  Cha Orl  
yoop  Dal Hou Phx Por      Chi LAC Mia NOP Sac

Ties (rounded to integer) are listed. Some of the 'closest' guesses are worse than some of the 'furthest'. The best guess at the Mavs is off by 8, while the worst on Sac and Mem are only off by 4.

Posted: **Sat Dec 03, 2016 10:39 pm**

RMSE is a traditional, common measure of model fit. It's been utilized in many fields. You can find answers of why squared errors are used pretty easily on google and it's taught (or at least I hope it is) in early statistics classes.

http://stats.stackexchange.com/question ... ared-error

I actually don't know what the error distribution is, but I don't think it's symmetrical, at least not for teams at the two extreme ends. So I don't know if RMSE is particularly needed for NBA wins, but the errors are squared for good reasons.

Posted: **Sun Dec 04, 2016 1:19 am**

We know light and sound fall off with the square of distance, gravitational velocity is proportional to time squared, etc. But what does that say about average errors of predictions?

When you're creating formulas, you might square the errors to reduce the outliers the most. But you can do that with an exponent of 1.4 or 2.2 or anything else.

The tradition may be simply that old timers with pen and paper could calculate a square root; a 1.4 root is much harder or impractical. I don't believe 2 is a magic number here, any more than 1/2 is . They just distort the measure in opposite ways.

It's also hard to use RMSE in a sentence. Can you verb it like you can with 'average'?

If you get the Lakers exactly right, and I am off by 4; and you're off by 16 on the Spurs, where my error is 12; then both our avg errors = 8.0
But in RMSE, you're off by (0+256)^1/2 = 16, while mine is (16+144)^1/2 = 12.65
Is that fair? Were you really more off than I was?

Posted: **Sun Dec 04, 2016 12:45 pm**

The 'news' today is that we all shrank our projected errors, from -.03 to -.28 (ATCt); yet regressed pyth'16 got bigger by .11, so we all gained relative to that.

Code: Select all

RyRi   4.34   .93      BaDo   4.85   .42
sbs.   4.48   .79      ATCt   4.90   .37
amp5   4.50   .77      GK5.   4.93   .34
sndi   4.53   .74      taco   4.99   .28
ncsB   4.59   .68      jg34   5.00   .27
shad   4.63   .64      cali   5.06   .21
Mike   4.64   .63      ncsD   5.08   .19
trzu   4.67   .60      Crow   5.10   .17
lnqi   4.75   .52      16py   5.27   .00
vegas  4.77   .50      nrfo   5.33  -.06
kmed   4.79   .48      yoop   5.50  -.23

And relative to shadow's weighted power ratings:

Code: Select all

RyRi   4.44   .52      shad   4.87    .09
sbs.   4.48   .48      ncsB   4.87    .09
sndi   4.54   .42      trzu   4.92    .04
GK5.   4.55   .41     16PyR   4.96    .00
taco   4.57   .39      jg34   5.01   -.05
Mike   4.65   .31     vegas   5.05   -.09
amp5   4.71   .25      yoop   5.19   -.23
ATCt   4.80   .16      cali   5.22   -.26
ncsD   4.81   .15      BaDo   5.34   -.38
kmed   4.84   .12      Crow   5.57   -.61
lnqi   4.86   .10      nrfo   5.72   -.76

Posted: **Tue Dec 06, 2016 6:29 pm**

If someone had just averaged our 20 entries, they'd be vying for the lead here.

Code: Select all

. avg error  impr      . avg error  impr
RyRi   4.09   .25      ATCt   4.69   .21
avg    4.16            GK5.   4.69   .24
amp5   4.21   .29      cali   4.72   .34
sbs.   4.24   .24      Crow   4.72   .38
Mike   4.30   .34      taco   4.84   .15
trzu   4.33   .34      jg34   4.86   .14
shad   4.34   .29      ncsD   4.94   .14
sndi   4.36   .17      nrfo   5.14   .19
ncsB   4.40   .19      16PyR  5.22   .05
kmed   4.42   .37      yoop   5.35   .15
BaDo   4.44   .41          
vegas  4.48   .29            
lnqi   4.48   .27

Good to serious improvement for every participant, in just one day.

Posted: **Thu Dec 08, 2016 12:28 pm**

Today we see the smallest error to date, for the leader and perhaps for most. And everyone is now better than regressed 2016 Pyth.

Code: Select all

RyRi   3.89   1.48      lnqi   4.42   .96
amp5   4.04   1.33      BaDo   4.42   .95
avg    4.08   1.30      kmed   4.43   .94
ncsB   4.09   1.29      taco   4.62   .76
shad   4.15   1.23      Crow   4.63   .75
sbs.   4.15   1.23      ncsD   4.68   .69
trzu   4.26   1.12      cali   4.71   .67
Mike   4.32   1.05      jg34   4.72   .65
sndi   4.36   1.02      GK5.   4.76   .61
vegas  4.37   1.01      nrfo   4.80   .57
ATCt   4.39    .99      yoop   5.26   .12
.                       16PyR  5.38   .00

That 2nd number is how far you are ahead of '16 Pythag-Regressed.

EDIT: Relative to latest shadow weighted power ratings:

Code: Select all

RyRi   4.31   1.52      taco   4.77  1.06
avg    4.33   1.49      lnqi   4.81  1.02
sbs.   4.35   1.47      ncsD   4.81  1.01
amp5   4.39   1.44      BaDo   4.87   .96
ncsB   4.49   1.34      jg34   4.91   .92
sndi   4.59   1.24      cali   4.98   .85
Mike   4.59   1.23     vegas   4.99   .84
shad   4.63   1.20      yoop   5.11   .71
ATCt   4.63   1.20      Crow   5.18   .65
kmed   4.66   1.17      nrfo   5.60   .23
GK5.   4.66   1.17     16PyR   5.83   .00
trzu   4.71   1.12

viewtopic.php?f=2&t=9228&p=28376#p28376

Posted: **Tue Dec 13, 2016 12:32 pm**

Greatest separation to date between the leader and the rest:

Code: Select all

Abs Err Avg  >16py      Abs Err Avg  >16py
RyRi   3.95   1.74      sndi   4.57   1.12
amp5   4.19   1.50      Crow   4.59   1.10
avg    4.20   1.49      kmed   4.59   1.10
trzu   4.20   1.49      nrfo   4.62   1.07
sbs.   4.22   1.47      taco   4.73    .95
ATCt   4.40   1.29      jg34   4.90    .79
lnqi   4.44   1.25      ncsD   4.99    .69
vegas  4.47   1.22      cali   5.01    .68
shad   4.47   1.21      GK5.   5.14    .55
ncsB   4.48   1.21      yoop   5.16    .53
BaDo   4.48   1.21     16PyR   5.69    .00
Mike   4.51   1.18

... and ...

Code: Select all

.      RMSE  >16py             RMSE >16py
RyRi   4.94   1.56      taco   5.45  1.04
amp5   5.01   1.49      kmed   5.46  1.04
avg    5.03   1.47      ATCt   5.54   .95
sbs.   5.15   1.34      ncsD   5.58   .91
lnqi   5.20   1.30      nrfo   5.74   .76
ncsB   5.21   1.28      cali   5.77   .72
trzu   5.32   1.17      Crow   5.81   .68
sndi   5.35   1.15      jg34   5.92   .57
vegas  5.35   1.14      GK5.   5.95   .54
Mike   5.36   1.14      yoop   6.17   .33
BaDo   5.41   1.09     16PyR   6.50   .00
shad   5.42   1.08

Posted: **Sun Dec 18, 2016 3:25 pm**

Again we've sunk to new lows.

Code: Select all

RyRi   3.53   2.02      BaDo   4.19   1.37
avg    3.76   1.79      sndi   4.19   1.37
trzu   3.79   1.77      kmed   4.22   1.34
amp5   3.84   1.72      Mike   4.26   1.29
ncsB   3.86   1.70      taco   4.29   1.27
ATCt   3.87   1.68      ncsD   4.39   1.17
sbs.   3.90   1.66      Crow   4.42   1.13
lnqi   3.93   1.63      jg34   4.46   1.09
vegas  4.02   1.53      nrfo   4.48   1.08
shad   4.06   1.50      cali   4.48   1.07
.                       yoop   4.78    .78
.                       GK5.   4.82    .74
.                      16PyR   5.56    0

RyanRiot has made one post to this forum. None of his predictions were the high or the low one for any team. But he guessed on the "right" side on most teams:

Code: Select all

tm    avg    proj    err   RyRi  -dif
Min   38.7   32.2   -6.5   36.7   2.0
Atl   42.5   38.9   -3.6   40.6   1.9
Mil   31.9   42.8   10.9   33.8   1.9
Cha   41.5   43.6    2.1   42.9   1.4
Det   41.2   43.4    2.2   42.5   1.3

Mia   36.2   34.7   -1.5   35.0   1.2
NYK   35.7   39.0    3.3   36.8   1.1
Phx   28.5   26.7   -1.8   27.7    .8
Uta   48.3   50.0    1.7   48.9    .6
Tor   50.4   57.2    6.8   51.0    .6

Bos   50.7   44.8   -5.9   50.2    .5
Phl   24.1   24.5     .4   24.6    .3
Ind   40.2   37.9   -2.3   40.0    .2
Brk   25.3   25.6     .3   25.5    .2
Sac   36.2   36.0    -.2   36.1    .1

Was   41.4   37.3   -4.1   41.3    .1
SAS   56.9   56.8    -.1   57.0   -.1
Mem   39.5   43.8    4.3   39.4   -.1
Cle   56.8   56.7    -.1   56.9   -.1
Dal   37.8   28.6   -9.2   38.0   -.2

Den   36.1   36.0    -.1   36.3   -.2
GSW   68.8   63.6   -5.2   69.1   -.3
Chi   39.3   41.4    2.1   38.9   -.4
Okl   45.6   43.6   -2.0   46.0   -.4
LAL   23.7   30.7    7.0   23.1   -.6

Por   44.0   37.1   -6.9   44.7   -.7
Hou   46.8   56.6    9.9   45.8  -1.0
LAC   51.2   56.8    5.6   50.2  -1.0
Orl   35.4   32.8   -2.6   36.4  -1.0
NOP   35.2   30.8   -4.4   36.3  -1.1

These are ranked by how much he's beating the field (avg). He's actually worse than avg on 14 of 30, but by smaller amounts.
He has the best (or tied) guess on about 7 teams; closest he comes to worst guess looks like Orl and Por, where he's within 1 or so.

Posted: **Sun Dec 18, 2016 9:13 pm**

Ryan Riot.

Is it a Friday Riot by chance? In Tampa? On Radio? Or am I barking up wrong tree? If I am, are you a musician in Canada or Switzerland?

I hope you will share more about your method or yourself. If you feel like it.

Posted: **Mon Dec 19, 2016 12:08 pm**

Having been reminded that 538 also has a projection ongoing, here we are relative to that forecast:

Code: Select all

RyRi   3.61   2.46      ATCt   4.16   1.91
trzu   3.67   2.40      shad   4.20   1.87
sbs.   3.77   2.29      nrfo   4.20   1.87
avg    3.79   2.27      cali   4.20   1.87
amp5   3.80   2.27      Crow   4.23   1.83
lnqi   3.82   2.24      ncsD   4.33   1.73
BaDo   3.87   2.20      sndi   4.40   1.67
vegas  3.87   2.19      jg34   4.66   1.41
ncsB   3.93   2.13      taco   4.80   1.27
Mike   4.00   2.07      yoop   4.87   1.20
kmed   4.08   1.98      GK5.   4.92   1.15
.    vvvvvvvvvv        16PyR   6.07    .00

A closer field and more separation from 2015-16 pyth-regressed.
I believe these are yesterday's numbers, so they can be compared to the previous chart.

APBRmetrics

2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion

Re: 2016-17 Team Win Projection Contest / Discussion