Sports and Fractals
Konstantinos Pelechrinis
@kpelechrinis
Sports …
…and fractals?
…and fractals?
Self similarity
• A self similar complex system or object exhibits the same statistical
properties over different scales
• Scale invariance à ∃Δ: f λx = 𝜆) 𝑓 𝑥 , ∀𝜆
Power laws
K. J. Hsu and A. Hsu, “Self-similarity of the “1/f noise” called music”,
in PNAS, vol 88, pp 3507-3509, April 1991
Coastline paradox
Why sports though?
• Several spatial sports data
• Player mobility, shot charts etc.
• The complexity of similar objects can be captured through the notion of
fractal dimensionality
With C(r) being the fraction of pairs of points from a set S that have distance smaller or equal to r,
S behaves like a fractal with intrinsic fractal dimension D in the range of scales r1 to r2, iff:
𝐶 𝑟 ∝ 𝑟1
, 𝑟2 ≤ 𝑟 ≤ 𝑟4
Fractal dimension
Fractal dimension = 0.48 Fractal dimension = 1.61 Fractal dimension = 1.94
Box counting
Fractal dimensionality and … dancing
Dance Fractal dimensionality
Rumba 1.36
Cha cha 1.24
Salsa 1.28
Merengue 1.16
Bachata 1.21
Salsa
Merengue
M. Tatlier and R. Suvak, “How fractal is dancing?”, Chaos, Solitons and Fractals 36 (2008) 1019-1027
Fractal dimensionality and … dancing
Dance Fractal dimensionality
Rumba 1.36
Cha cha 1.24
Salsa 1.28
Merengue 1.16
Bachata 1.21
Salsa
Merengue
M. Tatlier and R. Suvak, “How fractal is dancing?”, Chaos, Solitons and Fractals 36 (2008) 1019-1027
3pt shots & floor utilization dimensionality
3PT distance changes
Record breaking
5
10
15
20
25
1980 1990 2000 2010
Year
3PTAttemptsperGame
3pt shots & floor utilization dimensionality
3PT distance changes
Record breaking
5
10
15
20
25
1980 1990 2000 2010
Year
3PTAttemptsperGame
0.000
0.025
0.050
0.075
0 10 20 30 40
Distance (feet)
Density
0.000
0.025
0.050
0.075
0 10 20 30 40
Distance (feet)
Density
0.000
0.025
0.050
0.075
0.100
0 10 20 30 40
Distance (feet)
Density
0.000
0.025
0.050
0.075
0 10 20 30 40
Distance (feet)
Density
0.000
0.025
0.050
0.075
0.100
0 10 20 30 40
Distance (feet)
Density
3pt shots & floor utilization dimensionality
3PT distance changes
Record breaking
5
10
15
20
25
1980 1990 2000 2010
Year
3PTAttemptsperGame
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2015-16 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2014-15 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2014-15 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2014-15 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2015-16 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
3pt shots & floor utilization dimensionality
3PT distance changes
Record breaking
5
10
15
20
25
1980 1990 2000 2010
Year
3PTAttemptsperGame
0.000
0.025
0.050
0.075
0 10 20 30 40
Distance (feet)
Density
0.000
0.025
0.050
0.075
0 10 20 30 40
Distance (feet)
Density
0.000
0.025
0.050
0.075
0.100
0 10 20 30 40
Distance (feet)
Density
0.000
0.025
0.050
0.075
0 10 20 30 40
Distance (feet)
Density
0.000
0.025
0.050
0.075
0.100
0 10 20 30 40
Distance (feet)
Density
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2015-16 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2014-15 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2014-15 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2014-15 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
0.0
0.1
0.2
0.3
0.4
2PT 3PT
2015-16 Regular Season
FieldGoalPercentages
Position
Crest
Left
Right
Made Shot
Missed Shot
Above the Break 3
In The Paint (Non-RA)
Mid-Range
Restricted Area
Backcourt
Left Corner 3
Right Corner 3
Sampled Shot Chart
Floor utilization dimensionality
Isn’t this something we already know?!?!
Yes, but this shows that fractal dimensionality can be useful
Fractal versatility index
• Objective: quantify the spatial distribution of a player’s shot choices
• Proxy for versatility
Made Shot
Missed Shot
2014-15 Shot Chart
Giannis Antetokounmpo
NDSL@Pitt
FD = 0.72
Shot chart fractal dimensionality
1.42 1.38 1.11 0.96
0.57 0.31 1.27 1.44
Shot chart fractal dimensionality
1.42 1.38 1.11 0.96
0.57 0.31 1.27 1.44
Shot chart fractal dimensionality
1.42 1.38 1.11 0.96
0.57 0.31 1.27 1.44
Radius of gyration
140.6 166.3 140.7 125.6
25.4 36.2 113.2 83.2
Fractal Versatility Index
0.4
0.8
1.2
1.6
50 100 150 200
Radius of Gyration
FractalDimension
factor(clus)
1
2
3
4
5
6
7
8
9
10
Fractal Versatility Index
0.4
0.8
1.2
1.6
50 100 150 200
Radius of Gyration
FractalDimension
factor(clus)
1
2
3
4
5
6
7
8
9
10
S. Curry, J. Harden etc.
Fractal Versatility Index
0.4
0.8
1.2
1.6
50 100 150 200
Radius of Gyration
FractalDimension
factor(clus)
1
2
3
4
5
6
7
8
9
10
S. Curry, J. Harden etc.
D. Jordan, T. Duncan etc.
Fractal dimension and ball movement
• Fractal dimensionality of the ball’s trajectory is an one number description of ball
movement
• Ball movement can be intuitively thought of as a self-similar process
• Evaluation? Possibly manual but time-consuming
• Indirect evaluation: good ball movement should produce successful possessions
FBD Evaluation
• Logistic regression with DV the success (made FG) or not (missed FG, TO)
of a possession and FBD of the possession as the only IV
FBD Evaluation
FBD = 0.62, Made FG2
FBD = 1.33, Missed FG3
Soccer analytics
and fractals
• Can we use similar notions
to explain efficiency?
• For efficiency measures we
need an expected goals
model
Team efficiency
𝑂𝐸7 =	
𝐺7,: − 𝐸 𝐺7,:
𝐺7,:
𝐷𝐸7 =
𝐺7,= − 𝐸 𝐺7,=
𝐺7,=
CHI
CO
COL
DAL
DC
HOU
LA
MTL
NE
NYCFC
NYRB
OCSC
PHI
POR
RSL
SEA
SJ
SKC
TOR
VAN
-60
-30
0
30
60
-20 0 20
Offensive Efficiency (%)
DefensiveEfficiency(%)
Teams
a
a
a
a
Complete
Defensive
Need Work
Offensive
Fractal dimension
0.00
0.25
0.50
0.75
1.00
0.4 0.8 1.2 1.6
Team Fractal Dimension
ECDF
Top 50-th percentile à 6.7% offensive efficiency
Bottom 50-th percentile à -9.3% offensive efficiency
Fractal dimension
Fractal Dimension = 0.4
Goal
Miss
Fractal Dimension = 1.42
Goal
Miss
FD = 0.41 – Offensive efficiency = 19.4% FD = 1.43 – Offensive efficiency = -4.3%
More possibilities
• SportVU data
• PITCHf/x data
• Pitcher’s fractal dimension (?)
• Can it provide any new knowledge that we do not know already ?
• NextGen NFL stats
• Fractal dimension of successfully run routes (?)
Sources
• Perl code:
http://www.cs.cmu.edu/afs/cs.cmu.edu/user/christos/www/SRC/FracDim
-20001026.tar.gz
• R package: https://cran.r-
project.org/web/packages/fractaldim/fractaldim.pdf
• Matlab: https://www.mathworks.com/matlabcentral/fileexchange/13063-
boxcount
GLASC 2017

GLASC 2017

  • 1.
    Sports and Fractals KonstantinosPelechrinis @kpelechrinis
  • 2.
  • 3.
  • 4.
  • 5.
    Self similarity • Aself similar complex system or object exhibits the same statistical properties over different scales • Scale invariance à ∃Δ: f λx = 𝜆) 𝑓 𝑥 , ∀𝜆 Power laws K. J. Hsu and A. Hsu, “Self-similarity of the “1/f noise” called music”, in PNAS, vol 88, pp 3507-3509, April 1991 Coastline paradox
  • 6.
    Why sports though? •Several spatial sports data • Player mobility, shot charts etc. • The complexity of similar objects can be captured through the notion of fractal dimensionality With C(r) being the fraction of pairs of points from a set S that have distance smaller or equal to r, S behaves like a fractal with intrinsic fractal dimension D in the range of scales r1 to r2, iff: 𝐶 𝑟 ∝ 𝑟1 , 𝑟2 ≤ 𝑟 ≤ 𝑟4
  • 7.
    Fractal dimension Fractal dimension= 0.48 Fractal dimension = 1.61 Fractal dimension = 1.94
  • 8.
  • 9.
    Fractal dimensionality and… dancing Dance Fractal dimensionality Rumba 1.36 Cha cha 1.24 Salsa 1.28 Merengue 1.16 Bachata 1.21 Salsa Merengue M. Tatlier and R. Suvak, “How fractal is dancing?”, Chaos, Solitons and Fractals 36 (2008) 1019-1027
  • 10.
    Fractal dimensionality and… dancing Dance Fractal dimensionality Rumba 1.36 Cha cha 1.24 Salsa 1.28 Merengue 1.16 Bachata 1.21 Salsa Merengue M. Tatlier and R. Suvak, “How fractal is dancing?”, Chaos, Solitons and Fractals 36 (2008) 1019-1027
  • 12.
    3pt shots &floor utilization dimensionality 3PT distance changes Record breaking 5 10 15 20 25 1980 1990 2000 2010 Year 3PTAttemptsperGame
  • 13.
    3pt shots &floor utilization dimensionality 3PT distance changes Record breaking 5 10 15 20 25 1980 1990 2000 2010 Year 3PTAttemptsperGame 0.000 0.025 0.050 0.075 0 10 20 30 40 Distance (feet) Density 0.000 0.025 0.050 0.075 0 10 20 30 40 Distance (feet) Density 0.000 0.025 0.050 0.075 0.100 0 10 20 30 40 Distance (feet) Density 0.000 0.025 0.050 0.075 0 10 20 30 40 Distance (feet) Density 0.000 0.025 0.050 0.075 0.100 0 10 20 30 40 Distance (feet) Density
  • 14.
    3pt shots &floor utilization dimensionality 3PT distance changes Record breaking 5 10 15 20 25 1980 1990 2000 2010 Year 3PTAttemptsperGame 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2015-16 Regular Season FieldGoalPercentages Position Crest Left Right 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2014-15 Regular Season FieldGoalPercentages Position Crest Left Right 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2014-15 Regular Season FieldGoalPercentages Position Crest Left Right 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2014-15 Regular Season FieldGoalPercentages Position Crest Left Right 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2015-16 Regular Season FieldGoalPercentages Position Crest Left Right
  • 15.
    3pt shots &floor utilization dimensionality 3PT distance changes Record breaking 5 10 15 20 25 1980 1990 2000 2010 Year 3PTAttemptsperGame 0.000 0.025 0.050 0.075 0 10 20 30 40 Distance (feet) Density 0.000 0.025 0.050 0.075 0 10 20 30 40 Distance (feet) Density 0.000 0.025 0.050 0.075 0.100 0 10 20 30 40 Distance (feet) Density 0.000 0.025 0.050 0.075 0 10 20 30 40 Distance (feet) Density 0.000 0.025 0.050 0.075 0.100 0 10 20 30 40 Distance (feet) Density 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2015-16 Regular Season FieldGoalPercentages Position Crest Left Right 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2014-15 Regular Season FieldGoalPercentages Position Crest Left Right 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2014-15 Regular Season FieldGoalPercentages Position Crest Left Right 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2014-15 Regular Season FieldGoalPercentages Position Crest Left Right 0.0 0.1 0.2 0.3 0.4 2PT 3PT 2015-16 Regular Season FieldGoalPercentages Position Crest Left Right Made Shot Missed Shot Above the Break 3 In The Paint (Non-RA) Mid-Range Restricted Area Backcourt Left Corner 3 Right Corner 3 Sampled Shot Chart
  • 16.
    Floor utilization dimensionality Isn’tthis something we already know?!?! Yes, but this shows that fractal dimensionality can be useful
  • 17.
    Fractal versatility index •Objective: quantify the spatial distribution of a player’s shot choices • Proxy for versatility Made Shot Missed Shot 2014-15 Shot Chart Giannis Antetokounmpo NDSL@Pitt FD = 0.72
  • 18.
    Shot chart fractaldimensionality 1.42 1.38 1.11 0.96 0.57 0.31 1.27 1.44
  • 19.
    Shot chart fractaldimensionality 1.42 1.38 1.11 0.96 0.57 0.31 1.27 1.44
  • 20.
    Shot chart fractaldimensionality 1.42 1.38 1.11 0.96 0.57 0.31 1.27 1.44
  • 21.
    Radius of gyration 140.6166.3 140.7 125.6 25.4 36.2 113.2 83.2
  • 22.
    Fractal Versatility Index 0.4 0.8 1.2 1.6 50100 150 200 Radius of Gyration FractalDimension factor(clus) 1 2 3 4 5 6 7 8 9 10
  • 23.
    Fractal Versatility Index 0.4 0.8 1.2 1.6 50100 150 200 Radius of Gyration FractalDimension factor(clus) 1 2 3 4 5 6 7 8 9 10 S. Curry, J. Harden etc.
  • 24.
    Fractal Versatility Index 0.4 0.8 1.2 1.6 50100 150 200 Radius of Gyration FractalDimension factor(clus) 1 2 3 4 5 6 7 8 9 10 S. Curry, J. Harden etc. D. Jordan, T. Duncan etc.
  • 25.
    Fractal dimension andball movement • Fractal dimensionality of the ball’s trajectory is an one number description of ball movement • Ball movement can be intuitively thought of as a self-similar process • Evaluation? Possibly manual but time-consuming • Indirect evaluation: good ball movement should produce successful possessions
  • 26.
    FBD Evaluation • Logisticregression with DV the success (made FG) or not (missed FG, TO) of a possession and FBD of the possession as the only IV
  • 27.
    FBD Evaluation FBD =0.62, Made FG2 FBD = 1.33, Missed FG3
  • 28.
    Soccer analytics and fractals •Can we use similar notions to explain efficiency? • For efficiency measures we need an expected goals model
  • 29.
    Team efficiency 𝑂𝐸7 = 𝐺7,:− 𝐸 𝐺7,: 𝐺7,: 𝐷𝐸7 = 𝐺7,= − 𝐸 𝐺7,= 𝐺7,= CHI CO COL DAL DC HOU LA MTL NE NYCFC NYRB OCSC PHI POR RSL SEA SJ SKC TOR VAN -60 -30 0 30 60 -20 0 20 Offensive Efficiency (%) DefensiveEfficiency(%) Teams a a a a Complete Defensive Need Work Offensive
  • 30.
    Fractal dimension 0.00 0.25 0.50 0.75 1.00 0.4 0.81.2 1.6 Team Fractal Dimension ECDF Top 50-th percentile à 6.7% offensive efficiency Bottom 50-th percentile à -9.3% offensive efficiency
  • 31.
    Fractal dimension Fractal Dimension= 0.4 Goal Miss Fractal Dimension = 1.42 Goal Miss FD = 0.41 – Offensive efficiency = 19.4% FD = 1.43 – Offensive efficiency = -4.3%
  • 32.
    More possibilities • SportVUdata • PITCHf/x data • Pitcher’s fractal dimension (?) • Can it provide any new knowledge that we do not know already ? • NextGen NFL stats • Fractal dimension of successfully run routes (?)
  • 33.
    Sources • Perl code: http://www.cs.cmu.edu/afs/cs.cmu.edu/user/christos/www/SRC/FracDim -20001026.tar.gz •R package: https://cran.r- project.org/web/packages/fractaldim/fractaldim.pdf • Matlab: https://www.mathworks.com/matlabcentral/fileexchange/13063- boxcount