In [2]:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

from mplsoccer.pitch import Pitch

from sklearn.cluster import KMeans

In [17]:

#import data
df = pd.read_csv('kmeanstutorial.csv')

In [18]:

df.head()

Out[18]:

	Unnamed: 0	ball_receipt_outcome	ball_recovery_recovery_failure	block_deflection	carry_end_location	clearance_aerial_won	counterpress	dribble_outcome	dribble_overrun	duel_outcome	...	shot_statsbomb_xg	shot_technique	shot_type	substitution_outcome	substitution_replacement	tactics	team	timestamp	type	under_pressure
0	0	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	{'formation': 442, 'lineup': [{'player': {'id'...	France	00:00:00.000	Starting XI	NaN
1	1	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	{'formation': 433, 'lineup': [{'player': {'id'...	Croatia	00:00:00.000	Starting XI	NaN
2	2	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	France	00:00:00.000	Half Start	NaN
3	3	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	Croatia	00:00:00.000	Half Start	NaN
4	4	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	Croatia	00:00:00.000	Half Start	NaN

5 rows × 74 columns

In [19]:

df.columns

Out[19]:

Index(['Unnamed: 0', 'ball_receipt_outcome', 'ball_recovery_recovery_failure',
       'block_deflection', 'carry_end_location', 'clearance_aerial_won',
       'counterpress', 'dribble_outcome', 'dribble_overrun', 'duel_outcome',
       'duel_type', 'duration', 'foul_committed_advantage',
       'foul_committed_card', 'foul_committed_penalty', 'foul_committed_type',
       'foul_won_advantage', 'foul_won_defensive', 'goalkeeper_body_part',
       'goalkeeper_end_location', 'goalkeeper_outcome', 'goalkeeper_position',
       'goalkeeper_technique', 'goalkeeper_type', 'id', 'index',
       'injury_stoppage_in_chain', 'interception_outcome', 'location',
       'match_id', 'minute', 'pass_aerial_won', 'pass_angle',
       'pass_assisted_shot_id', 'pass_backheel', 'pass_body_part',
       'pass_cross', 'pass_cut_back', 'pass_deflected', 'pass_end_location',
       'pass_goal_assist', 'pass_height', 'pass_length', 'pass_outcome',
       'pass_recipient', 'pass_shot_assist', 'pass_switch', 'pass_type',
       'period', 'play_pattern', 'player', 'position', 'possession',
       'possession_team', 'related_events', 'second', 'shot_aerial_won',
       'shot_body_part', 'shot_deflected', 'shot_end_location',
       'shot_first_time', 'shot_freeze_frame', 'shot_key_pass_id',
       'shot_outcome', 'shot_statsbomb_xg', 'shot_technique', 'shot_type',
       'substitution_outcome', 'substitution_replacement', 'tactics', 'team',
       'timestamp', 'type', 'under_pressure'],
      dtype='object')

In [20]:

df = df[['team','type','location','pass_end_location']]

In [21]:

df.head()

Out[21]:

	team	type	location	pass_end_location
0	France	Starting XI	NaN	NaN
1	Croatia	Starting XI	NaN	NaN
2	France	Half Start	NaN	NaN
3	Croatia	Half Start	NaN	NaN
4	Croatia	Half Start	NaN	NaN

In [22]:

df = df[(df['team']=='France')&(df['type']=='Pass')].reset_index()

In [23]:

df.head()

Out[23]:

	index	team	type	location	pass_end_location
0	11	France	Pass	[48.0, 50.0]	[48.0, 60.0]
1	24	France	Pass	[49.0, 80.0]	[46.0, 61.0]
2	25	France	Pass	[65.0, 64.0]	[66.0, 69.0]
3	28	France	Pass	[63.0, 73.0]	[65.0, 79.0]
4	29	France	Pass	[58.0, 79.0]	[26.0, 69.0]

In [24]:

df.location.dtype

Out[24]:

dtype('O')

In [25]:

df[['x','y']] = df.location.str.split(expand=True)
df[['endX','endY']] = df.pass_end_location.str.split(expand=True)

In [26]:

df.head()

Out[26]:

	index	team	type	location	pass_end_location	x	y	endX	endY
0	11	France	Pass	[48.0, 50.0]	[48.0, 60.0]	[48.0,	50.0]	[48.0,	60.0]
1	24	France	Pass	[49.0, 80.0]	[46.0, 61.0]	[49.0,	80.0]	[46.0,	61.0]
2	25	France	Pass	[65.0, 64.0]	[66.0, 69.0]	[65.0,	64.0]	[66.0,	69.0]
3	28	France	Pass	[63.0, 73.0]	[65.0, 79.0]	[63.0,	73.0]	[65.0,	79.0]
4	29	France	Pass	[58.0, 79.0]	[26.0, 69.0]	[58.0,	79.0]	[26.0,	69.0]

In [27]:

df['x'] = df.loc[:,'x'] = df.x.map(lambda x: x[1:-1]).astype(float)
df['y'] = df.loc[:,'y'] = df.y.map(lambda x: x[0:-1]).astype(float)
df['endX'] = df.loc[:,'endX'] = df.endX.map(lambda x: x[1:-1]).astype(float)
df['endY'] = df.loc[:,'endY'] = df.endY.map(lambda x: x[0:-1]).astype(float)
df = df.drop(['location','pass_end_location'],axis=1)

In [28]:

df.head()

Out[28]:

	index	team	type	x	y	endX	endY
0	11	France	Pass	48.0	50.0	48.0	60.0
1	24	France	Pass	49.0	80.0	46.0	61.0
2	25	France	Pass	65.0	64.0	66.0	69.0
3	28	France	Pass	63.0	73.0	65.0	79.0
4	29	France	Pass	58.0	79.0	26.0	69.0

In [ ]:

#implement the kmeans

In [29]:

X = np.array(df[['x','y','endX','endY']])
kmeans = KMeans(n_clusters = 10,random_state=100)
kmeans.fit(X)
df['cluster'] = kmeans.predict(X)

In [30]:

df.head()

Out[30]:

	index	team	type	x	y	endX	endY
0	11	France	Pass	48.0	50.0	48.0	60.0
1	24	France	Pass	49.0	80.0	46.0	61.0
2	25	France	Pass	65.0	64.0	66.0	69.0
3	28	France	Pass	63.0	73.0	65.0	79.0
4	29	France	Pass	58.0	79.0	26.0	69.0

In [31]:

df.cluster.value_counts()

Out[31]:

0    41
1    40
6    38
8    31
3    31
7    28
2    26
9    25
4    18
5    14
Name: cluster, dtype: int64

In [34]:

fig, ax = plt.subplots(figsize=(10,10))
fig.set_facecolor('#38383b')
ax.patch.set_facecolor('#38383b')

pitch = Pitch(pitch_type='statsbomb',orientation='horizontal',
             pitch_color='#38383b',line_color='white',figsize=(10,10),
             constrained_layout=False,tight_layout=True,view='full')

pitch.draw(ax=ax)

for x in range(len(df['cluster'])):
    
    if df['cluster'][x] ==0:
        pitch.lines(xstart=df['x'][x],ystart=df['y'][x],xend=df['endX'][x],yend=df['endY'][x],
                   color='#74c69d',lw=3,zorder=2,comet=True,ax=ax)
        
    if df['cluster'][x] ==5:
        pitch.lines(xstart=df['x'][x],ystart=df['y'][x],xend=df['endX'][x],yend=df['endY'][x],
                   color='#add8e6',lw=3,zorder=2,comet=True,ax=ax)

In [ ]: