In [1]:
from mplsoccer.pitch import Pitch
import os
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from adjustText import adjust_text
import fuzzymatcher
import seaborn as sns
import matplotlib as mpl
%config InlineBackend.figure_format='retina'
In [2]:
WYSCOUT = os.path.join('..', 'data', 'wyscout')
STATSBOMB = os.path.join('..', 'data', 'statsbomb')
Load data
In [3]:
df_wyscout_event = pd.read_parquet(os.path.join(WYSCOUT, 'event_overlap.parquet'))
df_statsbomb_event = pd.read_parquet(os.path.join(STATSBOMB, 'event.parquet'))
df_wyscout_match = pd.read_parquet(os.path.join(WYSCOUT, 'match.parquet'))
df_wyscout_match_overlap = pd.read_parquet(os.path.join(WYSCOUT, 'match_overlap.parquet'))
df_wyscout_player = pd.read_parquet(os.path.join(WYSCOUT, 'player.parquet'))
df_statsbomb_match = pd.read_parquet(os.path.join(STATSBOMB, 'match.parquet'))
df_statsbomb_tactic = pd.read_parquet(os.path.join(STATSBOMB, 'tactic.parquet'))
Number of passes for starting players from StatsBomb data
In [4]:
#df_start = df_statsbomb_event.loc[df_statsbomb_event.type_name=='Starting XI', ['id']]
#df_start = df_start.merge(df_statsbomb_tactic, how='left')
#df_start = df_start[['match_id', 'player_id']]
#df_pass = df_statsbomb_event[df_statsbomb_event.type_name == 'Pass'].copy()
#df_pass_start = df_start.merge(df_pass, on=['match_id', 'player_id'], how='inner')
#df_pass_start.groupby(['match_id','player_id']).type_name.count().mean()
# 42.148323828812025
Games per competition
In [5]:
df_statsbomb_match.competition_name.value_counts()
Out[5]:
La Liga 452 FA Women's Super League 194 FIFA World Cup 64 Women's World Cup 52 NWSL 36 Premier League 32 Champions League 14 Name: competition_name, dtype: int64
In [6]:
df_statsbomb_match.groupby(['competition_name', 'season_name']).match_id.count()
Out[6]:
competition_name season_name
Champions League 2003/2004 1
2004/2005 1
2006/2007 1
2008/2009 1
2009/2010 1
2010/2011 1
2011/2012 1
2012/2013 1
2013/2014 1
2014/2015 1
2015/2016 1
2016/2017 1
2017/2018 1
2018/2019 1
FA Women's Super League 2018/2019 106
2019/2020 88
FIFA World Cup 2018 64
La Liga 2004/2005 7
2005/2006 17
2006/2007 26
2007/2008 28
2008/2009 31
2009/2010 35
2010/2011 33
2011/2012 37
2012/2013 32
2013/2014 31
2014/2015 38
2015/2016 33
2016/2017 34
2017/2018 36
2018/2019 34
NWSL 2018 36
Premier League 2003/2004 32
Women's World Cup 2019 52
Name: match_id, dtype: int64
% Shots belonging to Barcelona
In [7]:
num_barca_shots = ((df_statsbomb_event.team_name=='Barcelona') & (df_statsbomb_event.type_name=='Shot')).sum()
num_total_shots = (df_statsbomb_event.type_name=='Shot').sum()
print('Proportion of shots taken by Barcelona in StatsBomb data:', round(num_barca_shots/num_total_shots * 100, 1))
Proportion of shots taken by Barcelona in StatsBomb data: 33.3
Number shots relating to Messi
In [8]:
num_messi_shots = ((df_statsbomb_event.player_name == 'Lionel Andrés Messi Cuccittini') &
(df_statsbomb_event.type_name=='Shot')).sum()
In [9]:
print('Number of Messi shots in the StatsBomb dataset', num_messi_shots)
print('Number of total shots in the StatsBomb dataset', num_total_shots)
Number of Messi shots in the StatsBomb dataset 2031 Number of total shots in the StatsBomb dataset 21876
Combine wyscout match
In [10]:
df_wyscout_match = pd.concat([df_wyscout_match, df_wyscout_match_overlap])
Games in the Wyscout data
In [11]:
print(len(df_wyscout_match))
1941
In [12]:
df_wyscout_match.groupby(['season_name', 'competition_name']).season_name.count()
Out[12]:
season_name competition_name
2016 UEFA Euro 51
2017/2018 Bundesliga 306
La Liga 380
Ligue 1 380
Premier League 380
Serie A 380
2018 FIFA World Cup 64
Name: season_name, dtype: int64
In [13]:
df_statsbomb_shots = df_statsbomb_event[(df_statsbomb_event.type_name=='Shot')&
(df_statsbomb_event.sub_type_name!='Penalty')].dropna(how='all', axis=1).copy()
df_wyscout_shots = df_wyscout_event[((df_wyscout_event.eventName=='Shot')|
(df_wyscout_event.subEventName=='Free kick shot')) &
(df_wyscout_event['matchPeriod']!='P')].copy()
Subset StatsBomb overlap
In [14]:
mask_sb_overlap = ((df_statsbomb_match.competition_name == 'FIFA World Cup')|
((df_statsbomb_match.competition_name == 'La Liga') & (df_statsbomb_match.season_name == '2017/2018')))
overlap_id = df_statsbomb_match.loc[mask_sb_overlap, 'match_id']
df_statsbomb_event = df_statsbomb_event[df_statsbomb_event.match_id.isin(overlap_id)].copy()
Number of non-penalty shots in the Wyscout data
In [15]:
mask_shot_wy = (((df_wyscout_event.eventName=='Shot') | (df_wyscout_event.subEventName=='Free kick shot')) &
(df_wyscout_event['matchPeriod']!='P'))
print('Number of non-penalty shots:', mask_shot_wy.sum())
Number of non-penalty shots: 2420
Number of no-penalty shots in the StatsBomb data
In [16]:
mask_shot_sb = (df_statsbomb_event.type_name == 'Shot') & (df_statsbomb_event.sub_type_name!='Penalty')
print('Number of non-penalty shots - StatsBomb:', mask_shot_sb.sum())
Number of non-penalty shots - StatsBomb: 2604
Add match_id from other dataset
In [17]:
df_id1 = df_wyscout_match[['match_id', 'home_team_name', 'away_team_name', 'home_score', 'away_score', 'kick_off']].copy()
df_id2 = df_statsbomb_match[['match_id', 'home_team_name', 'away_team_name', 'home_score', 'away_score', 'kick_off']].copy()
df_id1['kick_off'] = pd.to_datetime(df_id1.kick_off.dt.date)
df_id2['kick_off'] = pd.to_datetime(df_id2.kick_off.dt.date)
df_overlap = df_id1.merge(df_id2, on=['home_team_name', 'away_team_name', 'home_score', 'kick_off'], how='inner',
suffixes=['_wyscout', '_statsbomb'])
df_overlap = df_overlap[['match_id_wyscout', 'match_id_statsbomb']]
df_statsbomb_event = df_statsbomb_event.merge(df_overlap, left_on='match_id', right_on='match_id_statsbomb', how='left')
df_wyscout_event = df_wyscout_event.merge(df_overlap, left_on='match_id', right_on='match_id_wyscout', how='left')
Add team name to Wyscout
In [18]:
df_team1 = df_wyscout_match[['home_team_id', 'home_team_name']].copy()
df_team1.rename({'home_team_id': 'team_id','home_team_name': 'team_name'}, axis=1, inplace=True)
df_team2 = df_wyscout_match[['away_team_id', 'away_team_name']].copy()
df_team2.rename({'away_team_id': 'team_id','away_team_name': 'team_name'}, axis=1, inplace=True)
df_teams = pd.concat([df_team1, df_team2]).drop_duplicates()
df_wyscout_event = df_wyscout_event.merge(df_teams, on='team_id', how='left')
Add player name to Wyscout. I am going to seperate out the last part of the name, the first part of the name and anything inbetween. This may not be technically accurate, but it's easier to join
In [19]:
df_wyscout_player['fullName'] = (df_wyscout_player.firstName + ' ' + df_wyscout_player.lastName).str.strip()
player_name_series = df_wyscout_player.fullName.str.split(' ')
df_wyscout_player['firstName'] = player_name_series.apply(lambda x: x[0] if isinstance(x, list) else None)
df_wyscout_player['middleName'] = player_name_series.apply(lambda x: ' '.join(x[1:-1]) if isinstance(x, list) else None)
df_wyscout_player['lastName'] = player_name_series.apply(lambda x: x[-1] if isinstance(x, list) else None)
df_wyscout_player['middleName'] = df_wyscout_player['middleName'].str.strip()
df_wyscout_player['Name'] = ((df_wyscout_player['firstName'] + ' ' + df_wyscout_player['middleName']).str.strip()
+ ' ' + df_wyscout_player['lastName'])
df_wyscout_event = df_wyscout_event.merge(df_wyscout_player, how='left')
Add player name to StatsBomb
In [20]:
player_name_series = df_statsbomb_event.player_name.str.strip().str.replace(pat='"', repl="'").str.split(' ')
df_statsbomb_event['firstName'] = player_name_series.apply(lambda x: x[0] if isinstance(x, list) else None)
df_statsbomb_event['middleName'] = player_name_series.apply(lambda x: ' '.join(x[1:-1]) if isinstance(x, list) else None)
df_statsbomb_event['middleName'] = df_statsbomb_event['middleName'].str.strip()
df_statsbomb_event['lastName'] = player_name_series.apply(lambda x: x[-1] if isinstance(x, list) else None)
df_statsbomb_event['Name'] = ((df_statsbomb_event['firstName'] + ' ' + df_statsbomb_event['middleName']).str.strip()
+ ' ' + df_statsbomb_event['lastName'])
Subset non-penalty shots
In [21]:
df_statsbomb_shots = df_statsbomb_event[(df_statsbomb_event.type_name=='Shot')&
(df_statsbomb_event.sub_type_name!='Penalty')].dropna(how='all', axis=1).copy()
df_wyscout_shots = df_wyscout_event[((df_wyscout_event.eventName=='Shot')|
(df_wyscout_event.subEventName=='Free kick shot')) &
(df_wyscout_event['matchPeriod']!='P')].copy()
In [22]:
print('Number of StatsBomb shots:', len(df_statsbomb_shots))
print('Number of Wyscout shots:', len(df_wyscout_shots))
Number of StatsBomb shots: 2604 Number of Wyscout shots: 2420
Add consistent period (in wyscout it's matchPeriod=1H/2H/E1/E2 in StatsBomb its period=1/2/3/4)
In [23]:
df_wyscout_shots['period'] = df_wyscout_shots['matchPeriod'].replace({'1H': 1, '2H': 2, 'E1': 3, 'E2': 4})
Add consistent seconds (eventSec) in Wyscout
In [24]:
df_statsbomb_shots['eventSec'] = ((df_statsbomb_shots.timestamp_minute * 60.) +
(df_statsbomb_shots.timestamp_second) +
(df_statsbomb_shots.timestamp_millisecond/1000.))
Create a combined player id
In [25]:
# merge the two datasets: exact match on names and team name
p1 = df_wyscout_shots[['player_id', 'firstName', 'middleName', 'lastName', 'Name', 'team_name']].drop_duplicates()
p2 = df_statsbomb_shots[['player_id', 'firstName', 'middleName', 'lastName', 'Name', 'team_name']].drop_duplicates()
p3 = p1.merge(p2, how='outer', on=['firstName', 'middleName',
'lastName', 'Name', 'team_name'], suffixes=['_wyscout', '_statsbomb'])
matched = p3[(p3.player_id_wyscout.notnull() & p3.player_id_statsbomb.notnull())].copy()
matched = matched[['player_id_wyscout', 'player_id_statsbomb']].drop_duplicates()
not_matched_statsbomb = p2[~p2.player_id.isin(matched.player_id_statsbomb)].copy()
not_matched_wyscout = p1[~p1.player_id.isin(matched.player_id_wyscout)].copy()
not_matched_wyscout.rename({'player_id': 'player_id_wyscout'}, axis=1, inplace=True)
not_matched_statsbomb.rename({'player_id': 'player_id_statsbomb'}, axis=1, inplace=True)
# Then non exact match using fuzzy matcher
# Note U hand checked the best_match_score threshold (anything above zero was actually a match in this case)
fuzzy_match = fuzzymatcher.fuzzy_left_join(not_matched_wyscout,
not_matched_statsbomb,
left_on=['Name', 'team_name'],
right_on=['Name', 'team_name'],
left_id_col='player_id_wyscout',
right_id_col='player_id_statsbomb')
fuzzy_match = fuzzy_match.loc[fuzzy_match.best_match_score > 0.3, ['player_id_wyscout', 'player_id_statsbomb']]
# all matched
matched = pd.concat([matched, fuzzy_match])
# get not matched
mask_no_match_wy = ~not_matched_wyscout.player_id_wyscout.isin(matched.player_id_wyscout)
not_matched_wyscout = not_matched_wyscout.loc[mask_no_match_wy, ['player_id_wyscout']].copy()
not_matched_wyscout.drop_duplicates(inplace=True)
mask_no_match_sb = ~not_matched_statsbomb.player_id_statsbomb.isin(matched.player_id_statsbomb)
not_matched_statsbomb = not_matched_statsbomb.loc[mask_no_match_sb, ['player_id_statsbomb']].copy()
not_matched_statsbomb.drop_duplicates(inplace=True)
# combined the match/ no match and get a unique id from the index
df_matched_players = pd.concat([matched, not_matched_statsbomb, not_matched_wyscout])
df_matched_players.reset_index(drop=True, inplace=True)
df_matched_players.index.name = 'player_id_combined'
df_matched_players.reset_index(inplace=True)
Add the combined player id to the datasets
In [26]:
df_statsbomb_shots = df_statsbomb_shots.merge(df_matched_players, left_on='player_id', right_on='player_id_statsbomb')
df_wyscout_shots = df_wyscout_shots.merge(df_matched_players, left_on='player_id', right_on='player_id_wyscout')
# hand check merge has not introduced additional shots
print('Number of StatsBomb shots:', len(df_statsbomb_shots))
print('Number of Wyscout shots:', len(df_wyscout_shots))
Number of StatsBomb shots: 2604 Number of Wyscout shots: 2420
Setup the pitches
In [27]:
pitch_statsbomb = Pitch(orientation='vertical', view='half')
pitch_wyscout = Pitch(pitch_type='wyscout', orientation='vertical', view='half')
pitch_statsperform = Pitch(pitch_type='statsperform', orientation='vertical', view='half', line_zorder=3)
Add metadata
In [28]:
df_statsbomb_shots = df_statsbomb_shots.merge(df_statsbomb_match[['match_id', 'metadata_shot_fidelity_version']],
on='match_id', how='left', validate='m:1')
@lemonwatcher says:
coordinates denote the bottom-right corner of a cell in a grid, not an actual point, so 1,1 is the top-left square 'yard' of the field (in landscape), even though 0,0 is the true coordinate of the corner flag.
some matches have metadata like "xy_fidelity_version" : "2", which means the grid has higher granularity. In this case 0.1,0.1 is the top left cell. We appear to have elided this bit of the docs in the repo, which I will attempt to remedy.
So here we set to the middle of the grid
In [29]:
mask_fidelity = df_statsbomb_shots.metadata_shot_fidelity_version!='2'
df_statsbomb_shots.loc[mask_fidelity, 'x'] = df_statsbomb_shots.loc[mask_fidelity, 'x'] - 0.5
df_statsbomb_shots.loc[mask_fidelity, 'y'] = df_statsbomb_shots.loc[mask_fidelity, 'y'] - 0.5
mask_fidelity = df_statsbomb_shots.metadata_shot_fidelity_version=='2'
df_statsbomb_shots.loc[mask_fidelity, 'x'] = df_statsbomb_shots.loc[mask_fidelity, 'x'] - 0.05
df_statsbomb_shots.loc[mask_fidelity, 'y'] = df_statsbomb_shots.loc[mask_fidelity, 'y'] - 0.05
Standardise to the same coordinate system 105*68
In [30]:
# statsbomb
df_statsbomb_shots[['x', 'end_x']] = ((df_statsbomb_shots[['x', 'end_x']])
/ float(pitch_statsbomb.right) * 105.)
df_statsbomb_shots[['y', 'end_y']] = ((float(pitch_statsbomb.bottom) - df_statsbomb_shots[['y', 'end_y']])
/ float(pitch_statsbomb.bottom) * 68.)
# wyscout
df_wyscout_shots[['x', 'end_x']] = ((df_wyscout_shots[['x', 'end_x']])
/ float(pitch_wyscout.right) * 105.)
df_wyscout_shots[['y', 'end_y']] = ((float(pitch_wyscout.bottom) - df_wyscout_shots[['y', 'end_y']])
/ float(pitch_wyscout.bottom) * 68.)
Subset interesting variables
In [31]:
df_statsbomb_shots = df_statsbomb_shots[['id', 'match_id_wyscout', 'player_id_combined', 'eventSec', 'period',
'sub_type_name', 'technique_name', 'team_name', 'Name',
'x', 'y', 'end_x', 'end_y', 'end_z',
'body_part_name', 'outcome_name']].copy()
In [32]:
df_wyscout_shots = df_wyscout_shots[['id', 'match_id_wyscout', 'player_id_combined', 'eventSec', 'period',
'team_name', 'Name', 'x', 'y', 'end_x', 'end_y', 'left_foot',
'right_foot', 'other_body_part', 'goal']].copy()
Merge shots
In [33]:
# sort the dataframe as needed for a merge_asof
df_statsbomb_shots.sort_values('eventSec', inplace=True)
df_wyscout_shots.sort_values('eventSec', inplace=True)
# copy the time column for working out time difference
df_statsbomb_shots['eventSec_statsbomb'] = df_statsbomb_shots.eventSec
df_wyscout_shots['eventSec_wyscout'] = df_wyscout_shots.eventSec
# match to closest shot by time - allow up to 20 second difference
df_statsbomb_shots_merged = pd.merge_asof(df_statsbomb_shots, df_wyscout_shots, on='eventSec',
by=['match_id_wyscout', 'player_id_combined', 'period'],
suffixes=('_statsbomb', '_wyscout'),
tolerance=26, direction='nearest')
matched_shots = df_statsbomb_shots_merged[df_statsbomb_shots_merged.id_wyscout.notnull()].copy()
unmatched_shots = df_statsbomb_shots_merged[df_statsbomb_shots_merged.id_wyscout.isnull()].copy()
# where statsbomb shot matched to multiple wyscout shots deduplicate
matched_shots['time_diff'] = (abs(matched_shots['eventSec_statsbomb'] - matched_shots['eventSec_wyscout']))
matched_shots.sort_values('time_diff', inplace=True)
matched_shots_dedup = matched_shots.drop_duplicates('id_wyscout', keep='first')
# get the real matches, i.e. the first match for a shot
mask_real_match = matched_shots.id_statsbomb.isin(matched_shots_dedup.id_statsbomb)
df_matched_shots = matched_shots[mask_real_match].copy() # collect matched shots
# get the unmatched shots
df_wyscout_unmatched = df_wyscout_shots[~df_wyscout_shots.id.isin(df_matched_shots.id_wyscout)].copy()
df_statsbomb_unmatched = df_statsbomb_shots[~df_statsbomb_shots.id.isin(df_matched_shots.id_statsbomb)].copy()
# hand check merge has not introduced additional shots
print('Number of StatsBomb shots:', len(df_statsbomb_unmatched) + len(df_matched_shots))
print('Number of Wyscout shots:', len(df_wyscout_unmatched) + len(df_matched_shots))
Number of StatsBomb shots: 2604 Number of Wyscout shots: 2420
Second merge to catch a few more
In [34]:
match_second = pd.merge_asof(df_statsbomb_unmatched, df_wyscout_unmatched, on='eventSec',
by=['match_id_wyscout', 'player_id_combined', 'period'],
suffixes=('_statsbomb', '_wyscout'),
tolerance=26, direction='nearest')
df_matched_shots = pd.concat([df_matched_shots, match_second[match_second.id_wyscout.notnull()]])
df_wyscout_unmatched = df_wyscout_unmatched[~df_wyscout_unmatched.id.isin(df_matched_shots.id_wyscout)]
df_statsbomb_unmatched = df_statsbomb_unmatched[~df_statsbomb_unmatched.id.isin(df_matched_shots.id_statsbomb)]
# hand check merge has not introduced additional shots
print('Number of StatsBomb shots:', len(df_statsbomb_unmatched) + len(df_matched_shots))
print('Number of Wyscout shots:', len(df_wyscout_unmatched) + len(df_matched_shots))
Number of StatsBomb shots: 2604 Number of Wyscout shots: 2420
In [35]:
print('Number of matched shots:', len(df_matched_shots))
print('Number of unmatched Wyscout shots:', len(df_wyscout_unmatched))
print('Number of unmatched StatsBomb shots:', len(df_statsbomb_unmatched))
Number of matched shots: 2386 Number of unmatched Wyscout shots: 34 Number of unmatched StatsBomb shots: 218
Plot unmatched StatsBomb shots
In [36]:
plt.style.use('dark_background')
pitch = Pitch(pitch_type='statsperform', figsize=(8, 9), view='half', line_zorder=3,
constrained_layout=True, tight_layout=False, orientation='vertical')
fig, ax = pitch.draw()
pitch.scatter(df_wyscout_unmatched.x, df_wyscout_unmatched.y, ax=ax, c='#6ae4cd',
label=f'Not in StatsBomb open-data ({len(df_wyscout_unmatched)} shots)')
pitch.scatter(df_statsbomb_unmatched.x, df_statsbomb_unmatched.y, ax=ax, c='#e8d247',
label=f'Not in Wyscout open-data ({len(df_statsbomb_unmatched)} shots)')
ax.legend(fontsize=15, loc='lower center',)
fig.savefig(os.path.join('..', 'figures', '12_overlap_shot_differences.png'), bbox_inches = 'tight', pad_inches = 0.2)
Plot difference
In [37]:
# goal probabilities for all shots
pitch = Pitch(pitch_type='statsperform', figsize=(16, 9), line_zorder=3, pad_top=2, line_color='#C3C3C3')
mask_goal = df_statsbomb_shots.outcome_name=='Goal'
bin_statistic_sb = pitch.bin_statistic(df_statsbomb_shots.x,
df_statsbomb_shots.y,
mask_goal,
statistic='mean',
bins=(5, 5))
# goal probabilities for matched shots
mask_goal = df_matched_shots.outcome_name=='Goal'
bin_statistic_match = pitch.bin_statistic(df_matched_shots.x_statsbomb,
df_matched_shots.y_statsbomb,
mask_goal,
statistic='mean',
bins=(5, 5))
# goal probabilities for all shots
mask_goal = df_statsbomb_shots.outcome_name=='Goal'
bin_statistic_sb = pitch.bin_statistic(df_statsbomb_shots.x,
df_statsbomb_shots.y,
mask_goal,
statistic='mean',
bins=(10, 10))
# goal probabilities for matched shots
mask_goal = df_matched_shots.outcome_name=='Goal'
bin_statistic_match = pitch.bin_statistic(df_matched_shots.x_statsbomb,
df_matched_shots.y_statsbomb,
mask_goal,
statistic='mean',
bins=(10, 10))
# calculate difference, i.e. what would happen if only have wyscout shots
bin_statistic_sb['statistic'] = np.nan_to_num(bin_statistic_match['statistic']) - np.nan_to_num(bin_statistic_sb['statistic'])
fig, ax = pitch.draw()
pcm = pitch.heatmap(bin_statistic_sb, ax=ax, cmap='viridis', edgecolors='#22312b', zorder=1)
bin_statistic_sb['statistic'] = np.round(bin_statistic_sb['statistic'] * 100., 0).astype(np.int32)
mask = bin_statistic_sb['statistic'] > 1
bin_statistic_sb['statistic'] = bin_statistic_sb['statistic'].astype(str)
label = pitch.label_heatmap(bin_statistic_sb, color='white', fontsize=30, ax=ax, ha='center', va='center')
bin_statistic_sb['statistic'][~mask] = ''
label = pitch.label_heatmap(bin_statistic_sb, color='black', fontsize=30, ax=ax, ha='center', va='center')
ax.set_title('The percentage point difference in goal probabilities\nwhen removing the shots not in the Wyscout data',
fontsize=20)
fig.savefig(os.path.join('..', 'figures', '13_overlap_shot_differences_probabilities.png'),
bbox_inches = 'tight', pad_inches = 0.2)
In [38]:
df_statsbomb_unmatched.outcome_name.value_counts()
Out[38]:
Wayward 99 Blocked 55 Off T 42 Saved 21 Post 1 Name: outcome_name, dtype: int64
In [39]:
df_statsbomb_unmatched.body_part_name.value_counts() / len(df_statsbomb_unmatched.body_part_name)
Out[39]:
Head 0.490826 Right Foot 0.298165 Left Foot 0.188073 Other 0.022936 Name: body_part_name, dtype: float64
In [40]:
df_statsbomb_unmatched[df_statsbomb_unmatched.body_part_name != 'Head'].outcome_name.value_counts()
Out[40]:
Wayward 40 Blocked 38 Off T 20 Saved 13 Name: outcome_name, dtype: int64
In [41]:
df_statsbomb_unmatched[df_statsbomb_unmatched.body_part_name == 'Head'].outcome_name.value_counts()
Out[41]:
Wayward 59 Off T 22 Blocked 17 Saved 8 Post 1 Name: outcome_name, dtype: int64
In [42]:
wayward = df_statsbomb_unmatched[df_statsbomb_unmatched.outcome_name == 'Wayward']
blocked = df_statsbomb_unmatched[df_statsbomb_unmatched.outcome_name == 'Blocked']
off_target = df_statsbomb_unmatched[df_statsbomb_unmatched.outcome_name == 'Off T']
other = df_statsbomb_unmatched[~df_statsbomb_unmatched.outcome_name.isin(['Wayward', 'Blocked', 'Off T'])]
In [43]:
pitch_grid = Pitch(pitch_type='statsperform', figsize=(16, 18), view='half', line_zorder=3,
layout=(2, 2), constrained_layout=True, tight_layout=False, goal_type='box')
fig, ax = pitch_grid.draw()
ax = ax.ravel()
labels = ['Wayward', 'Blocked', 'Off Target', 'Other']
for i, df in enumerate([wayward, blocked, off_target, other]):
pitch_grid.lines(df.x, df.y, df.end_x, df.end_y, alpha=0.7,
ax=ax[i], color='#e8d247', comet=True, transparent=True, lw=3,
label=f'{labels[i]} ({len(df)} umatched shots)')
pitch_grid.scatter(df.end_x, df.end_y, ax=ax[i], s=100, c='#6ae4cd', alpha=0.7,
label='end location', zorder=2)
ax[i].legend(loc='upper center', fontsize=20)
fig.suptitle('Type of shot not in the Wyscout data', fontsize=25)
fig.savefig(os.path.join('..', 'figures', '14_overlap_shot_type_unmatched.png'), bbox_inches = 'tight', pad_inches = 0.2)
Average distance between StatsBomb and Wyscout shot locations
In [44]:
distance = ((df_matched_shots.x_statsbomb - df_matched_shots.x_wyscout)**2 +
(df_matched_shots.y_statsbomb - df_matched_shots.y_wyscout)**2)**0.5
In [45]:
x_diff = df_matched_shots.x_statsbomb - df_matched_shots.x_wyscout
y_diff = df_matched_shots.y_statsbomb - df_matched_shots.y_wyscout
In [46]:
pd.DataFrame({'x_diff': x_diff, 'y_diff': y_diff, 'distance': distance}).describe()
Out[46]:
| x_diff | y_diff | distance | |
|---|---|---|---|
| count | 2386.000000 | 2386.000000 | 2386.000000 |
| mean | 1.637380 | -0.249870 | 3.557034 |
| std | 3.673168 | 3.080878 | 3.615352 |
| min | -7.787500 | -17.935000 | 0.121989 |
| 25% | 0.043750 | -2.125000 | 1.839465 |
| 50% | 1.312500 | -0.297500 | 3.049718 |
| 75% | 2.843750 | 1.615000 | 4.525746 |
| max | 85.312500 | 30.855000 | 86.780607 |
In [47]:
mpl.rcParams.update(mpl.rcParamsDefault)
sns.set(style="white", palette="muted", color_codes=True)
fig, ax = plt.subplots(2, 2, figsize=(16, 9), constrained_layout=True)
sns.distplot(x_diff, ax=ax[0, 0])
sns.distplot(y_diff, ax=ax[0, 1])
sns.distplot(distance, ax=ax[1, 0])
plt.setp(ax, yticks=[])
ax[0, 0].set_xlim(-15, 15)
ax[0, 0].annotate('x location', (-9, 0.1), fontsize=20)
ax[0, 1].set_xlim(-15, 15)
ax[0, 1].annotate('y location', (-10, 0.08), fontsize=20)
ax[1, 0].set_xlim(-15, 15)
ax[1, 0].annotate('distance', (-9, 0.08), fontsize=20)
ax[1, 1].remove()
fig.suptitle('Differences between the shot locations. Wyscout vs StatsBomb open-data \n for the 100 game overlap', fontsize=20)
fig.savefig(os.path.join('..', 'figures', '15_overlap_difference_distribution.png'), bbox_inches = 'tight', pad_inches = 0.2)
Plots example match
In [48]:
mpl.rcParams.update(mpl.rcParamsDefault)
plt.style.use('ggplot')
all_ids = df_statsbomb_shots.match_id_wyscout.unique()
idx = 1
subset = df_matched_shots[df_matched_shots.match_id_wyscout == all_ids[idx]].copy()
subset_match = df_wyscout_match[df_wyscout_match.match_id == all_ids[idx]]
team1 = subset_match.home_team_name.values[0]
team2 = subset_match.away_team_name.values[0]
fig, ax = plt.subplots(1, 2, figsize=(16, 9), constrained_layout=True)
pitch_statsperform.draw(ax=ax[0])
pitch_statsperform.scatter(subset.x_statsbomb, subset.y_statsbomb, c=np.arange(subset.y_statsbomb.size),
cmap='tab20b', ax=ax[0], s=300)
ax[0].set_title('Statsbomb shots', fontsize=30, color='black')
pitch_statsperform.draw(ax=ax[1])
pitch_statsperform.scatter(subset.x_wyscout, subset.y_wyscout, c=np.arange(subset.y_statsbomb.size),
cmap='tab20b', ax=ax[1], s=300)
ax[1].set_title('Wyscout shots', fontsize=30, color='black')
fig.savefig(os.path.join('..', 'figures', '16_overlap_difference_example.png'), bbox_inches = 'tight', pad_inches = 0.2)
