[Football Analysis] 8. 패스 네트워크 시각화(Pass Network Visualization)

작성일자 : 2023-12-23

Ver 0.1.1

 

 

패스 네트워크

In [1]:

# 필요 패키지 불러오기
import os
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from collections import defaultdict
import matplotlib.pyplot as plt
# from src.plot_utils import draw_pitch

pd.set_option('display.max_rows', 30)

In [2]:

# 현대 디렉토리 확인
current_dir = os.getcwd()
current_dir

Out[2]:

'/Users/limjongjun/Desktop/JayJay/Growth/Python/soccer-analytics/Excercise'

In [3]:

# Data가 있는 디렉토리로 변경
new_dir = '/Users/limjongjun/Desktop/JayJay/Growth/Python/soccer-analytics'
os.chdir(new_dir)
print(current_dir)

/Users/limjongjun/Desktop/JayJay/Growth/Python/soccer-analytics/Excercise

In [4]:

from src.plot_utils import draw_pitch

경기 데이터 불러오기¶

• PL 18 Round Manchester City vs Tottenham

In [5]:

# Manshester City VS Tottenham Hotspur 경기 불러오기
match_id = 2499895
match_events = pd.read_pickle(f'data/refined_events/England/2499895.pkl') #pkl 파일을 Dataframe으로 불러오기

In [6]:

match_events.head()

Out[6]:

	match_id	event_id	period	time	team_id	team_name	player_id	player_name	event_type	sub_event_type	tags	start_x	start_y	end_x	end_y
0	2499895	215108367	1H	1.784	1625	Manchester City	8325	S. Agüero	Pass	Simple pass	[Accurate]	52.00	34.68	40.56	34.68
1	2499895	215108368	1H	3.324	1625	Manchester City	105339	Fernandinho	Pass	Simple pass	[Accurate]	40.56	34.68	29.12	8.16
2	2499895	215108369	1H	6.406	1625	Manchester City	8277	K. Walker	Pass	Simple pass	[Accurate]	29.12	8.16	44.72	10.20
3	2499895	215108370	1H	7.124	1625	Manchester City	38021	K. De Bruyne	Pass	Simple pass	[Accurate]	44.72	10.20	69.68	6.12
4	2499895	215108371	1H	8.676	1625	Manchester City	11066	R. Sterling	Pass	Simple pass	[Accurate]	69.68	6.12	58.24	12.24

선수별 노드 시각화¶

(1) 선수별 패스 집계

In [7]:

passes = match_events[match_events['event_type'] == 'Pass']
pass_summary = passes.groupby(['team_name', 'player_name'])[['start_x', 'start_y']].mean()
pass_summary['total_count'] = passes.groupby(['team_name', 'player_name'])['event_id'].count()
pass_summary

Out[7]:

		start_x	start_y	total_count
team_name	player_name
Manchester City	Bernardo Silva	66.768000	16.320000	5
	E. Mangala	31.675429	47.833143	35
	Ederson	10.511429	36.817143	28
	F. Delph	51.114074	58.039259	54
	Fernandinho	45.003636	35.050909	33
	Gabriel Jesus	63.310000	24.990000	8
	K. De Bruyne	60.019556	27.426667	45
	K. Walker	45.963922	10.800000	51
	L. Sané	68.675862	57.846897	29
	N. Otamendi	30.346667	23.852308	39
	P. Foden	62.660000	52.445000	8
	R. Sterling	67.897143	17.291429	35
	S. Agüero	69.120000	37.870769	13
	İ. Gündoğan	60.613333	41.113846	39
Tottenham Hotspur	C. Eriksen	59.938667	39.734667	30
	D. Alli	59.150000	37.173333	24
	D. Rose	48.204000	60.656000	40
	E. Dier	27.476800	20.672000	50
	E. Lamela	66.560000	31.620000	4
	H. Kane	71.066667	31.597333	15
	H. Lloris	9.460645	35.425806	31
	H. Winks	51.447500	38.080000	32
	J. Vertonghen	26.961509	50.743396	53
	K. Trippier	49.589091	11.822727	44
	M. Dembélé	47.027500	35.785000	32
	M. Sissoko	48.880000	12.240000	2
	Son Heung-Min	71.209412	23.600000	17

(2) 경기 전체 선수별 노드 시각화

In [8]:

team_name = 'Manchester City'
nodes = pass_summary.loc[team_name].reset_index()
nodes

Out[8]:

	player_name	start_x	start_y	total_count
0	Bernardo Silva	66.768000	16.320000	5
1	E. Mangala	31.675429	47.833143	35
2	Ederson	10.511429	36.817143	28
3	F. Delph	51.114074	58.039259	54
4	Fernandinho	45.003636	35.050909	33
5	Gabriel Jesus	63.310000	24.990000	8
6	K. De Bruyne	60.019556	27.426667	45
7	K. Walker	45.963922	10.800000	51
8	L. Sané	68.675862	57.846897	29
9	N. Otamendi	30.346667	23.852308	39
10	P. Foden	62.660000	52.445000	8
11	R. Sterling	67.897143	17.291429	35
12	S. Agüero	69.120000	37.870769	13
13	İ. Gündoğan	60.613333	41.113846	39

In [9]:

draw_pitch('white', 'black', size_x=18, size_y=12)

x = nodes['start_x']
y = nodes['start_y']
s = nodes['total_count'] * 50
plt.scatter(x, y, s=s, c='skyblue', edgecolors='black')

for i, node in nodes.iterrows():
    plt.annotate(node['player_name'], xy=(node['start_x'], node['start_y']), c='k', fontsize=15)

• 위 코드는 교체 선수까지 모두 포함한(= 출전 시간을 고려하지 않고) 지표이므로 객관적인 분석 어려움

(3) 동시 출전 선수 추출을 위한 phase 구분

In [10]:

team_events = match_events[match_events['team_name'] == team_name] # Manchester City 선수들로만 필터링
player_change_records = team_events[
    (team_events['event_type'] == 'Substitution') |
    (team_events['tags'].apply(lambda x: 'Red card' in x))
]
in_players = player_change_records[player_change_records['sub_event_type'] == 'Player in']['player_id'].tolist()

player_ids = [p for p in team_events['player_id'].unique() if not p in in_players]

period_durations = match_events.groupby('period')['time'].max()
phase_record_list = []
phase = 1

for period in period_durations.index:
    change_times = player_change_records[player_change_records['period'] == period]['time'].unique().tolist()
    change_times.append(period_durations[period])
    if 0 not in change_times:
        change_times = [0] + change_times

    for i in range(len(change_times[:-1])):
        moment_records = player_change_records[
            (player_change_records['period'] == period) &
            (player_change_records['time'] == change_times[i])
        ]

        for _, record in moment_records.iterrows():
            if record['sub_event_type'] == 'Player out' or record['event_type'] == 'Foul':
                player_ids.remove(record['player_id'])
            else:
                player_ids.append(record['player_id'])

        phase_record = {
            'phase': phase,
            'period': period,
            'start_time': change_times[i],
            'end_time': change_times[i+1],
            'duration': change_times[i+1] - change_times[i],
            'player_ids': player_ids.copy()
        }
        phase += 1

        phase_record_list.append(phase_record)

phase_records = pd.DataFrame(phase_record_list).set_index('phase')
phase_records

Out[10]:

	period	start_time	end_time	duration	player_ids
phase
1	1H	0.0	2817.335	2817.335	[8325, 105339, 8277, 38021, 11066, 70086, 8464...
2	2H	0.0	660.000	660.000	[8325, 105339, 8277, 38021, 11066, 70086, 8464...
3	2H	660.0	2220.000	1560.000	[105339, 8277, 38021, 11066, 70086, 8464, 7165...
4	2H	2220.0	2340.000	120.000	[105339, 8277, 38021, 11066, 70086, 8464, 7165...
5	2H	2340.0	2878.216	538.216	[105339, 8277, 38021, 11066, 70086, 8464, 7165...

(4) 특정 phase 내 선수별 패스 집계

In [11]:

phase = 1

period = phase_records.at[phase, 'period']
start_time = phase_records.at[phase, 'start_time']
end_time = phase_records.at[phase, 'end_time']

team_phase_events = team_events[
    (team_events['period'] == period) &
    (team_events['time'] >= start_time) &
    (team_events['time'] <= end_time)
].reset_index(drop=True)

passes = team_phase_events[team_phase_events['event_type'] == 'Pass']
nodes = passes.groupby('player_name')[['start_x', 'start_y']].mean()
nodes['total_count'] = passes.groupby('player_name')['event_id'].count()
nodes

Out[11]:

	start_x	start_y	total_count
player_name
E. Mangala	35.310476	50.125714	21
Ederson	10.583529	38.480000	17
F. Delph	57.315556	58.177778	36
Fernandinho	46.861176	36.840000	17
K. De Bruyne	56.160000	29.604286	28
K. Walker	41.779310	11.864828	29
L. Sané	69.472000	60.248000	20
N. Otamendi	30.963636	26.890909	22
R. Sterling	70.026667	15.957333	15
S. Agüero	65.614545	40.429091	11
İ. Gündoğan	61.651200	42.486400	25

(5) 특정 phase 선수별 노드 시각화

In [12]:

draw_pitch('white', 'black', size_x=18, size_y=12)

x = nodes['start_x']
y = nodes['start_y']
s = nodes['total_count'] * 50
plt.scatter(x, y, s=s, c='skyblue', edgecolors='black')

for player_name, node in nodes.iterrows():
    plt.annotate(player_name, xy=(node['start_x'], node['start_y']), c='k', fontsize=15)

선수간 간선 시각화¶

(1) 주는 선수와 받는 선수 구분이 있는 선수간 패스 횟수 산출

• 선수간 패스 횟수를 defaultdict에 저장

In [13]:

pass_count_dict = defaultdict(int)
pass_count_dict

Out[13]:

defaultdict(int, {})

In [14]:

pass_count_dict['A'] = 1
pass_count_dict['A']

Out[14]:

1

In [15]:

pass_count_dict['B']

Out[15]:

0

In [16]:

pass_count_dict = defaultdict(int)

for i in team_phase_events.index[:-2]:
    event = team_phase_events.loc[i]
    next_event = team_phase_events.loc[i+1]

    if event['event_type'] == 'Pass' and 'Accurate' in event['tags']:
        sender = event['player_name']
        receiver = next_event['player_name']
        pass_count_dict[(sender, receiver)] += 1

pass_count_dict

Out[16]:

defaultdict(int,
            {('S. Agüero', 'Fernandinho'): 2,
             ('Fernandinho', 'K. Walker'): 3,
             ('K. Walker', 'K. De Bruyne'): 6,
             ('K. De Bruyne', 'R. Sterling'): 3,
             ('R. Sterling', 'K. De Bruyne'): 1,
             ('N. Otamendi', 'K. Walker'): 4,
             ('K. Walker', 'S. Agüero'): 2,
             ('R. Sterling', 'K. Walker'): 3,
             ('K. Walker', 'Fernandinho'): 4,
             ('Fernandinho', 'S. Agüero'): 1,
             ('L. Sané', 'F. Delph'): 7,
             ('F. Delph', 'L. Sané'): 9,
             ('L. Sané', 'E. Mangala'): 2,
             ('E. Mangala', 'İ. Gündoğan'): 1,
             ('K. De Bruyne', 'S. Agüero'): 5,
             ('S. Agüero', 'K. De Bruyne'): 3,
             ('Fernandinho', 'N. Otamendi'): 1,
             ('N. Otamendi', 'K. De Bruyne'): 4,
             ('İ. Gündoğan', 'K. Walker'): 2,
             ('K. Walker', nan): 1,
             ('R. Sterling', 'Fernandinho'): 2,
             ('Fernandinho', 'E. Mangala'): 3,
             ('E. Mangala', 'S. Agüero'): 2,
             ('İ. Gündoğan', 'F. Delph'): 9,
             ('F. Delph', 'E. Mangala'): 5,
             ('E. Mangala', 'F. Delph'): 4,
             ('F. Delph', 'S. Agüero'): 1,
             ('Ederson', 'E. Mangala'): 2,
             ('E. Mangala', 'Ederson'): 6,
             ('Ederson', 'K. Walker'): 5,
             ('K. De Bruyne', nan): 1,
             ('K. Walker', 'İ. Gündoğan'): 2,
             ('N. Otamendi', 'İ. Gündoğan'): 1,
             ('İ. Gündoğan', 'L. Sané'): 3,
             ('F. Delph', 'İ. Gündoğan'): 8,
             ('İ. Gündoğan', 'Fernandinho'): 2,
             ('Fernandinho', 'İ. Gündoğan'): 2,
             ('K. De Bruyne', 'Fernandinho'): 1,
             ('K. Walker', 'R. Sterling'): 4,
             ('İ. Gündoğan', 'N. Otamendi'): 1,
             ('Ederson', 'F. Delph'): 3,
             ('F. Delph', 'K. De Bruyne'): 3,
             ('K. De Bruyne', 'L. Sané'): 4,
             ('L. Sané', 'S. Agüero'): 3,
             ('S. Agüero', 'F. Delph'): 1,
             ('F. Delph', 'Fernandinho'): 3,
             ('Fernandinho', 'R. Sterling'): 2,
             ('R. Sterling', 'İ. Gündoğan'): 1,
             ('İ. Gündoğan', 'R. Sterling'): 1,
             ('K. De Bruyne', 'F. Delph'): 2,
             ('L. Sané', 'İ. Gündoğan'): 2,
             ('N. Otamendi', 'L. Sané'): 1,
             ('E. Mangala', 'N. Otamendi'): 2,
             ('R. Sterling', 'L. Sané'): 2,
             ('Fernandinho', 'F. Delph'): 3,
             ('İ. Gündoğan', 'E. Mangala'): 2,
             ('Ederson', 'Fernandinho'): 1,
             ('N. Otamendi', 'Fernandinho'): 2,
             ('Fernandinho', 'L. Sané'): 1,
             ('N. Otamendi', 'Ederson'): 3,
             ('L. Sané', 'K. De Bruyne'): 2,
             ('K. De Bruyne', 'E. Mangala'): 1,
             ('K. De Bruyne', 'K. Walker'): 1,
             ('Ederson', 'K. De Bruyne'): 2,
             ('S. Agüero', 'İ. Gündoğan'): 1,
             ('İ. Gündoğan', 'S. Agüero'): 1,
             ('S. Agüero', 'L. Sané'): 1,
             ('K. De Bruyne', 'N. Otamendi'): 1,
             ('F. Delph', 'Ederson'): 2,
             ('İ. Gündoğan', 'K. De Bruyne'): 1,
             ('K. De Bruyne', 'İ. Gündoğan'): 2,
             ('E. Mangala', nan): 1,
             ('R. Sterling', 'N. Otamendi'): 2,
             ('N. Otamendi', 'F. Delph'): 2,
             ('K. Walker', nan): 1,
             ('R. Sterling', nan): 1,
             ('R. Sterling', 'S. Agüero'): 1,
             ('N. Otamendi', 'E. Mangala'): 1,
             ('K. Walker', 'Ederson'): 1,
             ('S. Agüero', 'R. Sterling'): 1,
             ('Ederson', 'S. Agüero'): 1,
             ('S. Agüero', nan): 1,
             ('K. Walker', 'N. Otamendi'): 2,
             ('N. Otamendi', 'R. Sterling'): 1,
             ('K. Walker', 'K. Walker'): 1,
             ('E. Mangala', nan): 1,
             ('K. Walker', 'E. Mangala'): 1,
             ('Ederson', 'N. Otamendi'): 1,
             ('K. De Bruyne', nan): 1})

• defaultdict 내부 정보를 DataFrame에 저장

In [17]:

pass_count_list = []

for pair, count in pass_count_dict.items():
    pass_count_list.append([pair[0], pair[1], count])

pass_count_df = pd.DataFrame(pass_count_list, columns=['sender', 'receiver', 'count'])
pass_count_df

Out[17]:

	sender	receiver	count
0	S. Agüero	Fernandinho	2
1	Fernandinho	K. Walker	3
2	K. Walker	K. De Bruyne	6
3	K. De Bruyne	R. Sterling	3
4	R. Sterling	K. De Bruyne	1
...	...	...	...
84	K. Walker	K. Walker	1
85	E. Mangala	NaN	1
86	K. Walker	E. Mangala	1
87	Ederson	N. Otamendi	1
88	K. De Bruyne	NaN	1

89 rows × 3 columns

(2) 주는 선수와 받는 선수 구분이 없는 선수간 패스 횟수 산출

In [18]:

# 강의 코드
# pass_count_dict = defaultdict(int)

# for i in team_phase_events.index[:-2]:
#     event = team_phase_events.loc[i]
#     next_event = team_phase_events.loc[i+1]

#     if event['event_type'] == 'Pass' and 'Accurate' in event['tags']:
#         player_pair = [event['player_name'], next_event['player_name']]
#         player1 = min(player_pair)
#         player2 = max(player_pair)
#         pass_count_dict[(player1, player2)] += 1

# pass_count_list = []

# for pair, count in pass_count_dict.items():
#     pass_count_list.append([pair[0], pair[1], count])

# edges = pd.DataFrame(pass_count_list, columns=['player1', 'player2', 'count'])
# edges

# 수정 코드
## player_name 열의 모든 값을 문자열로 변환
team_phase_events['player_name'] = team_phase_events['player_name'].astype(str)

for i in team_phase_events.index[:-2]:
    event = team_phase_events.loc[i]
    next_event = team_phase_events.loc[i+1]

    if event['event_type'] == 'Pass' and 'Accurate' in event['tags']:
        player_pair = [event['player_name'], next_event['player_name']]
        player_pair.sort()  # 선수 이름을 정렬하여 순서를 통일
        pass_count_dict[tuple(player_pair)] += 1

pass_count_list = []

for pair, count in pass_count_dict.items():
    pass_count_list.append([pair[0], pair[1], count])

edges = pd.DataFrame(pass_count_list, columns=['player1', 'player2', 'count'])

(3) 패스 횟수 데이터에 평균 패스 위치 병합하기

• Player 1 평균 패스 위치 병합

In [19]:

edges = pd.merge(edges, nodes[['start_x', 'start_y']], left_on='player1', right_index=True)
edges

Out[19]:

	player1	player2	count	start_x	start_y
0	S. Agüero	Fernandinho	2	65.614545	40.429091
15	S. Agüero	K. De Bruyne	3	65.614545	40.429091
44	S. Agüero	F. Delph	1	65.614545	40.429091
64	S. Agüero	İ. Gündoğan	3	65.614545	40.429091
66	S. Agüero	L. Sané	1	65.614545	40.429091
...	...	...	...	...	...
40	Ederson	F. Delph	8	10.583529	38.480000
56	Ederson	Fernandinho	2	10.583529	38.480000
63	Ederson	K. De Bruyne	4	10.583529	38.480000
80	Ederson	S. Agüero	2	10.583529	38.480000
87	Ederson	N. Otamendi	5	10.583529	38.480000

102 rows × 5 columns

• Player 2 평균 패스 위치 병합

In [20]:

edges = edges.rename(columns={'start_x': 'player1_x', 'start_y': 'player1_y'})
edges = pd.merge(edges, nodes[['start_x', 'start_y']], left_on='player2', right_index=True)
edges

Out[20]:

	player1	player2	count	player1_x	player1_y	start_x	start_y
0	S. Agüero	Fernandinho	2	65.614545	40.429091	46.861176	36.840000
8	K. Walker	Fernandinho	4	41.779310	11.864828	46.861176	36.840000
37	K. De Bruyne	Fernandinho	1	56.160000	29.604286	46.861176	36.840000
20	R. Sterling	Fernandinho	2	70.026667	15.957333	46.861176	36.840000
57	N. Otamendi	Fernandinho	2	30.963636	26.890909	46.861176	36.840000
...	...	...	...	...	...	...	...
27	Ederson	E. Mangala	2	10.583529	38.480000	35.310476	50.125714
78	K. Walker	Ederson	1	41.779310	11.864828	10.583529	38.480000
59	N. Otamendi	Ederson	3	30.963636	26.890909	10.583529	38.480000
68	F. Delph	Ederson	2	57.315556	58.177778	10.583529	38.480000
28	E. Mangala	Ederson	14	35.310476	50.125714	10.583529	38.480000

89 rows × 7 columns

In [21]:

edges = edges.rename(columns={'start_x': 'player2_x', 'start_y': 'player2_y'}).reset_index(drop=True)
edges

Out[21]:

	player1	player2	count	player1_x	player1_y	player2_x	player2_y
0	S. Agüero	Fernandinho	2	65.614545	40.429091	46.861176	36.840000
1	K. Walker	Fernandinho	4	41.779310	11.864828	46.861176	36.840000
2	K. De Bruyne	Fernandinho	1	56.160000	29.604286	46.861176	36.840000
3	R. Sterling	Fernandinho	2	70.026667	15.957333	46.861176	36.840000
4	N. Otamendi	Fernandinho	2	30.963636	26.890909	46.861176	36.840000
...	...	...	...	...	...	...	...
84	Ederson	E. Mangala	2	10.583529	38.480000	35.310476	50.125714
85	K. Walker	Ederson	1	41.779310	11.864828	10.583529	38.480000
86	N. Otamendi	Ederson	3	30.963636	26.890909	10.583529	38.480000
87	F. Delph	Ederson	2	57.315556	58.177778	10.583529	38.480000
88	E. Mangala	Ederson	14	35.310476	50.125714	10.583529	38.480000

89 rows × 7 columns

(4) 선수간 간선(edge) 시각화

In [22]:

draw_pitch('white', 'black', size_x=18, size_y=12)

x = nodes['start_x']
y = nodes['start_y']
s = nodes['total_count'] * 100
plt.scatter(x, y, s=s, c='skyblue', edgecolors='black', zorder=2)

for player_name, node in nodes.iterrows():
    plt.annotate(player_name, xy=(node['start_x'], node['start_y']), c='k', fontsize=15, zorder=3)

max_count = edges['count'].max()
for i, edge in edges.iterrows():
    alpha = edge['count'] / max_count * 0.7 + 0.3
    plt.plot(
        edge[['player1_x', 'player2_x']], edge[['player1_y', 'player2_y']],
        lw=edge['count'] * 2, color='grey', alpha=alpha, zorder=1
    )

# plt.savefig('img/pass_network.png', bbox_inches='tight')
plt.show()