March 21, 2017 rajiv shah RajivShah.com [email protected] github.com/rajshah4/ rajcs4 Basketball Analytics Using Motion Tracking Data
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
March 21, 2017
rajiv shah
RajivShah.com [email protected]
github.com/rajshah4/
rajcs4
Basketball Analytics Using Motion Tracking Data
What is motion data?
descriptive use
statistical models
deep learning
goals
huh?
SportVu system for motion capture
In 2013 - 82.7%
SportVu motion data for a play
Game video
SportVu motion data for a play
Tabular motion data
Finding the data . . .htt
ps://github.com
/rajshah4/B
ask
etb
allData
/
i. descriptive
stats
player movement
htt
p://p
roje
cts
.rajivsh
ah.com
/sportvu/E
DA_NBA_SportVu.h
tml
htt
p://p
roje
cts
.rajivsh
ah.com
/sportvu/E
DA_NBA_SportVu.h
tml
player movement
htt
p://p
roje
cts
.rajivsh
ah.com
/sportvu/E
DA_NBA_SportVu.h
tml
htt
p://w
ww.slo
ansp
ortsc
onfe
rence.com
/wp-c
onte
nt/uplo
ads/2013
/Accele
ration
%20in
%20th
e%
20NBA%
20Toward
s%20an%
20Alg
orith
mic
%20Taxonom
y%
20of
%20Bask
etb
all%
20Pla
ys.pdf
htt
p://w
ww.slo
ansp
ortsc
onfe
rence.com
/wp-c
onte
nt/uplo
ads/2013
/Accele
ration
%20in
%20th
e%
20NBA%
20Toward
s%20an%
20Alg
orith
mic
%20Taxonom
y%
20of
%20Bask
etb
all%
20Pla
ys.pdf
offensive three second violations
htt
p://w
ww.chro
nic
let.com
/im
age/2
017
/03/16/x
600_q65/D
R-L
ast
-Shot-Reaction-1-jpg.jpg
offensive three second violations
ii. modeling
steve
steve & chris
htt
p://w
ww.b
ask
etb
allanaly
ticsb
ook.com
/2015
/09/14/p
relim
inary
-inve
stig
ation-
into
-defe
nsive
-stretc
h/
htt
p://w
ww.slo
ansp
ortsc
onfe
rence.com
/wp-c
onte
nt/uplo
ads/
2014
/02/2
014
_SSAC_The-T
hre
e-D
imensions-
Of-Reboundin
g.p
df
dan & alex
htt
p://w
ww.slo
ansp
ortsc
onfe
rence.com
/wp-c
onte
nt/uplo
ads/2014
/06/P
oin
twise.p
df
yisong & patrick
htt
p://p
roje
cts
.yisongyue.com
/bballpre
dic
t/
dan & alex
htt
p://w
ww.n
ess
is.o
rg/n
ess
is13
/cerv
one.p
df
Spatial random e↵ect surfaces for made shot events
Parker
−2
0
2
4
Green
−2
−1
0
1
2
3
4
Duncan
−2
−1
0
1
2
3
Leonard
−2
−1
0
1
2
3
4
Ginobili
−2
0
2
4
Blair
−1
0
1
2
3
4
dan & alex
Spatial random e↵ect surfaces for pass eventsParker to Duncan
Passer surface
−1.5
−1.0
−0.5
0.0
0.5
1.0
1.5
Receiver surface
−1.0
−0.5
0.0
0.5
1.0
1.5
2.0
Duncan to ParkerPasser surface
−2
−1
0
1
2
Receiver surface
−2
−1
0
1
2
dan & alex
Putting it all together
21
3
5
4
I 1: ParkerI 2: JacksonI 3: GreenI 4: DuncanI 5: Diaw
Pass next:
E [X |pass] = (0.78)(0.02)
+ (1.08)(0.14)
+ (0.84)(0.37)
+ (0.85)(0.46)
= 0.87
P(pass) = 0.97
Shoot next:
E [X |shot] = (3.00)(0.18)
+ (0.18)(0.82)
= 0.69
P(shot) = 0.03
EPV:
(0.87)(0.97) + (0.69)(0.03)
= 0.86
dan & alex
htt
ps://github.com
/dcerv
one/E
PVDem
o
iii. deep learning
htt
p://w
ww.slo
ansp
ortsc
onfe
rence.com
/wp-c
onte
nt/uplo
ads/2016
/02/1536-
Cla
ssifyin
g-N
BA-O
ffensive
-Pla
ys-
Using-N
eura
l-Netw
ork
s.pdf
Deep Learning to Basketball TrajectoriesAuthors: Rajiv Shah & Rob Romijnders
Results: Recurrent neural network beats
feature rich models
Problem: Predict whether a three point shot will be made
htt
p://p
roje
cts
.rajivsh
ah.com
/sportvu/T
raj_
RNN.h
tml
htt
p://w
ww.yisongyue.com
/publications/nip
s2016
_traje
cto
ry.p
df
htt
p://w
ww.slo
ansp
ortsc
onfe
rence.com
/wp-c
onte
nt/uplo
ads/2017
/02/1690.p
df
more cameras & data
htt
p://w
ww.slo
ansp
ortsc
onfe
rence.com
/wp-c
onte
nt/uplo
ads/2017
/02/1595.p
df
Understand the trends in basketball analytics
progression in analytic techniques
code and data that you can dig in
widen your appreciation
whew . .
March 21, 2017
rajiv shah
RajivShah.com [email protected]
github.com/rajshah4/
rajcs4
Basketball Analytics Using Motion Tracking Data
Related Documents
