sportowe_wizualizacja/xGStatsbomb.ipynb at 0eb052eb7aa9ab9a03af5656b606554fb9dd06b2

%%time
!git clone https://github.com/statsbomb/open-data.git

Cloning into 'open-data'...
remote: Enumerating objects: 1088, done.[K
remote: Counting objects: 100% (1088/1088), done.[K
remote: Compressing objects: 100% (591/591), done.[K
remote: Total 9810 (delta 893), reused 674 (delta 479), pack-reused 8722[K
Receiving objects: 100% (9810/9810), 995.57 MiB | 14.28 MiB/s, done.
Resolving deltas: 100% (8640/8640), done.
Checking out files: 100% (1648/1648), done.
CPU times: user 548 ms, sys: 115 ms, total: 663 ms
Wall time: 2min 44s

#import all modules
import json
import os
import codecs
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.patches import Arc, Rectangle, ConnectionPatch
from matplotlib.offsetbox import  OffsetImage
from matplotlib.patches import Ellipse
from functools import reduce
import math

%%time
comp = ['FIFA World Cup','La Liga']
main_df = pd.DataFrame(data=None)
path_match = "/content/open-data/data/events/" #location for play by play events
for root, dirs, files in os.walk('/content/open-data/data/matches/'):
     for file in files:
        with open(os.path.join(root, file), "r") as auto:
            with codecs.open(root + str('/') + file,encoding='utf-8') as data_file:
                data = json.load(data_file)
                df = pd.DataFrame(data=None)
                df = pd.json_normalize(data, sep = "_")
            #for x in df.competition_country_name:
            #    if x == 'Spain':
            #        print(df.match_id)
            #print(df['competition_competition_name'])
            for i in range(len(df)):
                if df.iloc[i]['competition_competition_name'] in comp  :
                    match_no = df.iloc[i]['match_id'] #gets match with Spain as country
                    match_no = str(match_no) # from int to str 
                    #print('match list \n',match_no)
                    with codecs.open(path_match + match_no + str(r'.json'),encoding="utf8") as event_file: #open the respective file
                        df_match = json.load(event_file)
                        df_match2 = pd.DataFrame(data=None)
                        df_match2 = pd.json_normalize(df_match,sep="_")    
                        df_match2 =  df_match2[(df_match2['type_name'] == "Shot")]
                    main_df = main_df.append(df_match2,ignore_index=True,sort=False)                    
#print('total matches ',len(match_no)) 
print('Done')

Done
CPU times: user 6min 3s, sys: 1.3 s, total: 6min 4s
Wall time: 6min 4s

main_df.head()

	id	index	period	timestamp	minute	second	possession	duration	type_id	type_name	possession_team_id	possession_team_name	play_pattern_id	play_pattern_name	team_id	team_name	tactics_formation	tactics_lineup	related_events	location	player_id	player_name	position_id	position_name	pass_recipient_id	pass_recipient_name	pass_length	pass_angle	pass_height_id	pass_height_name	pass_end_location	pass_type_id	pass_type_name	pass_body_part_id	pass_body_part_name	carry_end_location	under_pressure	pass_outcome_id	pass_outcome_name	pass_aerial_won	...	substitution_outcome_id	substitution_outcome_name	substitution_replacement_id	substitution_replacement_name	shot_one_on_one	bad_behaviour_card_id	bad_behaviour_card_name	50_50_outcome_id	50_50_outcome_name	dribble_overrun	goalkeeper_punched_out	pass_miscommunication	block_deflection	pass_goal_assist	clearance_other	injury_stoppage_in_chain	shot_deflected	dribble_no_touch	pass_deflected	shot_saved_off_target	goalkeeper_shot_saved_off_target	ball_recovery_offensive	pass_straight	foul_committed_penalty	foul_won_penalty	block_save_block	shot_open_goal	goalkeeper_lost_out	goalkeeper_success_in_play	player_off_permanent	goalkeeper_shot_saved_to_post	shot_redirect	shot_saved_to_post	shot_follows_dribble	goalkeeper_success_out	half_start_late_video_start	goalkeeper_lost_in_play	goalkeeper_saved_to_post	pass_backheel	half_end_early_video_end
0	2f046b33-685c-4122-8af2-8ceadf56c83d	294	1	00:06:50.216	6	50	12	0.115400	16	Shot	217	Barcelona	4	From Throw In	217	Barcelona	NaN	NaN	[58295c63-1ffa-4e27-9258-818ea90c6b04, f514442...	[104.4, 41.8]	5503.0	Lionel Andrés Messi Cuccittini	17.0	Right Wing	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1	855d47fc-9017-4508-8b41-0275dfb4d755	962	1	00:22:27.038	22	27	38	2.046458	16	Shot	217	Barcelona	2	From Corner	217	Barcelona	NaN	NaN	[aec80f5c-807e-47ac-8c33-092c92b222d1]	[110.8, 35.8]	5470.0	Ivan Rakitić	10.0	Center Defensive Midfield	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
2	7c69fb86-c77d-463b-8f00-503e447492a4	1153	1	00:27:08.522	27	8	46	0.804175	16	Shot	217	Barcelona	2	From Corner	217	Barcelona	NaN	NaN	[350f13e2-16cc-449d-a72d-f7ccd571fc50, 662299b...	[109.9, 40.5]	5492.0	Samuel Yves Umtiti	5.0	Left Center Back	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	True	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
3	77ea8775-f9f4-4bf7-b3f9-7635ab861ab5	1254	1	00:30:13.151	30	13	59	0.380900	16	Shot	217	Barcelona	3	From Free Kick	217	Barcelona	NaN	NaN	[30b9d0e1-5eeb-4cb0-86ea-a6e8967893e2, ae620c7...	[90.0, 36.2]	5503.0	Lionel Andrés Messi Cuccittini	17.0	Right Wing	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
4	09c2667a-4827-4871-a70a-96adb1d73243	1381	1	00:33:19.875	33	19	63	0.222600	16	Shot	217	Barcelona	4	From Throw In	217	Barcelona	NaN	NaN	[19491e5f-dd7c-47a8-994d-b6aae0630b55, a81b342...	[97.3, 28.8]	6998.0	Rafael Alcântara do Nascimento	15.0	Left Center Midfield	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN

5 rows × 147 columns

"""Distance of shot location to centre of goal"""
def distFormula(coordinate):
  a =(math.sqrt(((coordinate.location[0] - 120)**2) + ((coordinate.location[1] - 36)**2))) 
  b =(math.sqrt(((coordinate.location[0] - 120)**2) + ((coordinate.location[1] - 44)**2))) 
  return ((a+b)/2)

""" near x y (nx,ny) (120,44)and far x y (fx,fy) (120,36)""" 
nx = 120
ny = 44
fx = 120
fy = 36

goalpostLength = 8
def shot_angle(points):
  len1 = (math.sqrt(((points.location[0] - nx)**2) + ((points.location[1] - ny)**2))) 
  len2 = (math.sqrt(((points.location[0] - fx)**2) + ((points.location[1] - fy)**2)))
  ang = (len1**2 + len2**2 - goalpostLength**2)/(2 * len1 * len2)
  if ang > 1:
    ang = 1
  elif ang < -1:
    ang = -1   
  angRad = math.acos(ang)
  return( (angRad * 180)/math.pi)

"""If shot was taken under Pressure?"""
def under_pressure(coordinate):
  if coordinate['under_pressure'] == True:
    return 1
  return 0

"""The Shot type Id"""
def shot_type(coordinate):
  if coordinate['shot_type_id'] == 61:
    return 1
  if coordinate['shot_type_id'] == 62:
    return 2
  if coordinate['shot_type_id'] == 87:
    return 3
  if coordinate['shot_type_id'] == 88:
    return 4
  return 5

"""The Shot Body part"""
def shot_body_part(coordinate):
  if coordinate['shot_body_part_id'] == 37:
    return 1
  if coordinate['shot_body_part_id'] == 38:
    return 2
  if coordinate['shot_body_part_id'] == 70:
    return 3
  return 4

"""The Shot Technique Id"""
def shot_technique(coordinate):
  if coordinate['shot_technique_id'] == 89:
    return 1
  if coordinate['shot_technique_id'] == 90:
    return 2
  if coordinate['shot_technique_id'] == 91:
    return 3
  if coordinate['shot_technique_id'] == 92:
    return 4
  if coordinate['shot_technique_id'] == 93:
    return 5
  if coordinate['shot_technique_id'] == 94:
    return 6
  return 7

"""If shot was taken first time?"""
def shot_first_time(coordinate):
  if coordinate['shot_first_time'] == True:
    return 1
  return 0

"""If shot was taken first time?"""
def shot_one_on_one(coordinate):
  if coordinate['shot_one_on_one'] == True:
    return 1
  return 0

main_df['Distance'] = main_df.apply(distFormula,axis = 1)
main_df['Angle'] = main_df.apply(shot_angle,axis = 1)
main_df['UnderPressure'] = main_df.apply(under_pressure,axis = 1)
main_df['ShotType'] = main_df.apply(shot_type,axis = 1)
main_df['ShotBodyPart'] = main_df.apply(shot_body_part,axis = 1)
main_df['ShotTechnique'] = main_df.apply(shot_technique,axis = 1)
main_df['ShotFirstTime'] = main_df.apply(shot_first_time,axis = 1)
main_df['ShotOneonOne']= main_df.apply(shot_one_on_one,axis = 1)

goals_lst = main_df[main_df['shot_outcome_id'] == 97].index.tolist()

#if shot is a goal 
main_df['isGoal'] = False
goals_lst
main_df.loc[main_df.index.isin(goals_lst),'isGoal'] = True

main_df[['location','Distance','Angle','UnderPressure','ShotType','ShotBodyPart','ShotTechnique','ShotFirstTime','ShotOneonOne','isGoal']]

	location	Distance	Angle	UnderPressure	ShotType	ShotBodyPart	ShotTechnique	ShotFirstTime	ShotOneonOne	isGoal
0	[104.4, 41.8]	16.198841	28.422114	0	3	4	5	1	0	False
1	[110.8, 35.8]	10.763067	40.465393	0	3	1	5	0	0	False
2	[109.9, 40.5]	10.873186	43.128076	1	3	1	5	0	0	False
3	[90.0, 36.2]	30.499043	14.956182	0	2	2	5	0	0	False
4	[97.3, 28.8]	25.566766	16.208386	0	3	4	5	0	0	False
...	...	...	...	...	...	...	...	...	...	...
12952	[111.0, 27.0]	15.981653	17.102729	0	3	2	5	0	0	False
12953	[114.0, 33.0]	9.619084	34.824489	0	3	2	5	0	0	True
12954	[107.0, 32.0]	15.646638	25.606661	0	3	2	5	0	0	False
12955	[97.0, 22.0]	29.376742	12.398277	0	3	4	5	0	0	False
12956	[109.0, 52.0]	16.508979	19.464104	0	3	4	5	0	0	False

12957 rows × 10 columns

#xG Model

from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
import xgboost as xgb
from sklearn import svm
from sklearn import linear_model

xgModel = main_df[['location','Distance','Angle','UnderPressure','ShotType','ShotBodyPart','ShotTechnique','ShotFirstTime','ShotOneonOne','isGoal']]

X_train,X_test,y_train,y_test = train_test_split(xgModel[['location','Distance','Angle','UnderPressure','ShotType','ShotBodyPart','ShotTechnique','ShotFirstTime','ShotOneonOne']],xgModel['isGoal'],test_size = 0.2,shuffle = True)

Logistic Regression Model

clf = LogisticRegression(random_state=0,max_iter = 5000).fit(X_train[['Distance','Angle','UnderPressure','ShotType','ShotBodyPart','ShotTechnique','ShotFirstTime','ShotOneonOne']], y_train)

#model weights
clf.coef_[0]

array([-0.06605534,  0.02878253, -0.60546461,  1.03559372,  0.18762927,
        0.01151648,  0.18877916,  0.57099517])

SGD Model

xG = clf.predict_proba(X_test[['Distance','Angle','UnderPressure','ShotType','ShotBodyPart','ShotTechnique','ShotFirstTime','ShotOneonOne']])[:,1]

#SGD
sgdclf = linear_model.SGDClassifier(loss='log', alpha = 0.17)
sgdclf.fit(X_train[['Distance','Angle','UnderPressure','ShotType','ShotBodyPart','ShotTechnique','ShotFirstTime','ShotOneonOne']], y_train)

SGDClassifier(alpha=0.17, average=False, class_weight=None,
              early_stopping=False, epsilon=0.1, eta0=0.0, fit_intercept=True,
              l1_ratio=0.15, learning_rate='optimal', loss='log', max_iter=1000,
              n_iter_no_change=5, n_jobs=None, penalty='l2', power_t=0.5,
              random_state=None, shuffle=True, tol=0.001,
              validation_fraction=0.1, verbose=0, warm_start=False)

Predict Shot Probability

# change model here sgcclf(SGD) or clf(LR)
xG = sgdclf.predict_proba(X_test[['Distance','Angle','UnderPressure','ShotType','ShotBodyPart','ShotTechnique','ShotFirstTime','ShotOneonOne']])[:,1]

X_test['xG'] = xG
#X_test.head()

	location	Distance	Angle	UnderPressure	ShotType	ShotBodyPart	ShotTechnique	ShotOneonOne	xG
2961	[92.2, 58.9]	33.779300	11.289656	0	3	4	5	0	0.035840
10990	[115.2, 45.8]	8.019390	43.348531	0	3	1	5	1	0.340789
12649	[115.0, 41.0]	6.451010	75.963757	0	3	1	5	0	0.453033
5264	[111.9, 32.5]	11.445052	31.472019	1	3	1	5	0	0.082841
9283	[111.6, 37.0]	9.696832	46.594546	1	3	1	5	0	0.135449

sortxg = X_test.sort_values(by = ['xG'],ascending=False)
sortxg

	location	Distance	Angle	UnderPressure	ShotType	ShotBodyPart	ShotTechnique	ShotFirstTime	ShotOneonOne	xG
4624	[119.3, 41.4]	4.068882	157.545469	0	3	4	5	1	0	0.955584
3818	[119.1, 42.6]	4.162706	139.499608	0	3	4	5	1	0	0.927110
6952	[118.4, 39.4]	4.313989	135.619868	0	3	4	5	1	0	0.918449
8311	[119.2, 37.0]	4.163095	134.820390	0	3	1	5	0	1	0.902713
12822	[119.0, 43.0]	4.242641	126.869898	0	3	4	5	1	0	0.897920
...	...	...	...	...	...	...	...	...	...	...
3937	[69.2, 77.6]	63.283107	5.832625	0	3	4	4	0	0	0.004453
2834	[81.1, 5.9]	51.817864	6.672792	0	2	4	5	0	0	0.003497
6385	[57.2, 34.0]	63.211517	7.223482	0	3	2	5	0	0	0.003241
11776	[62.0, 36.0]	58.274562	7.853313	1	3	2	5	0	0	0.002498
6659	[51.9, 43.4]	68.301760	6.706436	0	3	2	4	0	0	0.002257

2592 rows × 10 columns

import StatsbombPitch as sb
sb.sb_pitch("#195905","#faf0e6","horizontal","full")
plt.gca().invert_yaxis()
for i in range(len(sortxg)):
    xe = sortxg.iloc[i]['location'][0]
    ye = sortxg.iloc[i]['location'][1]
    
    if sortxg.iloc[i]['xG'] >= 0.75:
      g = plt.scatter(xe,ye,color="#ee3e32",edgecolors="none",zorder=10,alpha=1,s = 40 )
    elif sortxg.iloc[i]['xG'] < 0.75 and sortxg.iloc[i]['xG'] >=0.5:
      o = plt.scatter(xe,ye,color="#f68838",edgecolors="none",zorder=8,alpha=0.75,s = 30 )
    elif sortxg.iloc[i]['xG'] < 0.5 and sortxg.iloc[i]['xG'] >=0.25:
      a = plt.scatter(xe,ye,color="#fbb021",edgecolors="none",zorder=6,alpha=0.5,s = 20 )  
    else:
      b = plt.scatter(xe,ye,color="#1b8a5a",edgecolors="none",zorder=4,alpha=0.25,s = 10 )    
plt.axis('off')
plt.legend((g,o,a,b),('>=0.75','>=0.5','>=0.25','<0.25'),scatterpoints=1,loc=2,title = 'xG Value',fontsize='small', fancybox=True)
#plt.title('xG SGD model')
#plt.savefig('xgSGDmodel.png')
plt.show()

import StatsbombPitch as sb
sb.sb_pitch("#195905","#faf0e6","vertical","half")
#plt.gca().invert_xaxis()
for i in range(len(sortxg)):
    xe = sortxg.iloc[i]['location'][0]
    ye = sortxg.iloc[i]['location'][1]
    
    if sortxg.iloc[i]['xG'] >= 0.75:
      g = plt.scatter(ye,xe,color="#ee3e32",edgecolors="none",zorder=10,alpha=1,s = 40 )
    elif sortxg.iloc[i]['xG'] < 0.75 and sortxg.iloc[i]['xG'] >=0.5:
      o = plt.scatter(ye,xe,color="#f68838",edgecolors="none",zorder=8,alpha=0.75,s = 30 )
    elif sortxg.iloc[i]['xG'] < 0.5 and sortxg.iloc[i]['xG'] >=0.25:
      a = plt.scatter(ye,xe,color="#fbb021",edgecolors="none",zorder=6,alpha=0.5,s = 20 )  
    else:
      b = plt.scatter(ye,xe,color="#1b8a5a",edgecolors="none",zorder=4,alpha=0.25,s = 10 )    
plt.axis('off')
plt.legend((g,o,a,b),('>=0.75','>=0.5','>=0.25','<0.25'),scatterpoints=1,loc=3,title = 'xG Value',fontsize='small', fancybox=True,edgecolor = 'black',framealpha = 2
           )


#ax = plt.subplot()

#plt.savefig('MessiValverdeEraScatter.png')
plt.show()

#[['Distance','Angle','UnderPressure','ShotType','ShotBodyPart','ShotTechnique','ShotFirstTime','ShotOneonOne']]
clf.predict_proba(np.array([12.55,37.156,0,4,2,5,0,0]).reshape(1, -1))[:,1]

array([0.38105829])

233 KiB Raw Blame History Unescape Escape

233 KiB

Raw Blame History