Notebook

In [184]:

import os

import numpy as np
import pandas as pd
#from pandas.api.types import CategoricalDtype
import matplotlib.pyplot as plt
import seaborn as sns
import altair as alt

from sklearn.preprocessing import MinMaxScaler

from ipywidgets import interact, interactive, fixed, interact_manual
import ipywidgets as widgets

In [2]:

years = [2020]

In [3]:

data = pd.DataFrame()
for i in years:  
    i_data = pd.read_csv('https://github.com/guga31bb/nflfastR-data/blob/master/data/' \
                         'play_by_play_' + str(i) + '.csv.gz?raw=True',
                         compression='gzip', low_memory=False)
    data = data.append(i_data, sort=True)

In [4]:

data = data.drop(columns=data.columns[data.columns.str.contains('epa') | data.columns.str.contains('wp')])
data = data.drop(columns=data.columns[data.columns.str.contains('home') | data.columns.str.contains('away')])
data = data.drop(columns=data.columns[data.columns.str.contains('tackle') | data.columns.str.contains('fumble')])
data = data.drop(columns=data.columns[data.columns.str.contains('punt') | data.columns.str.contains('kick')])
data = data.drop(columns=data.columns[data.columns.str.contains('xyac') | data.columns.str.contains('prob')])
data = data.drop(columns=data.columns[data.columns.str.contains('stadium')])
data = data.drop(columns=data.columns[data.columns.str.contains('blocked')])
data = data.drop(columns=data.columns[data.columns.str.contains('extra')])
data = data.drop(columns=data.columns[data.columns.str.contains('drive')])
data = data.drop(columns=data.columns[data.columns.str.contains('fantasy')])
data = data.drop(columns=data.columns[data.columns.str.contains('field_goal')])
data = data.drop(columns=data.columns[data.columns.str.contains('jersey')])
data = data.drop(columns=data.columns[data.columns.str.contains('lateral')])
data = data.drop(columns=data.columns[data.columns.str.contains('pass_def')])
data = data.drop(columns=data.columns[data.columns.str.contains('qb_hit')])
data = data.drop(columns=data.columns[data.columns.str.contains('return')])
data = data.drop(columns=data.columns[data.columns.str.contains('run')])
data = data.drop(columns=data.columns[data.columns.str.contains('rush')])
data = data.drop(columns=data.columns[data.columns.str.contains('two')])
data = data.drop(columns=data.columns[data.columns.str.contains('_down_')])
data = data.drop(columns=data.columns[data.columns.str.contains('end_')])
data = data.drop(columns=data.columns[data.columns.str.contains('_post')])
data = data.drop(columns=data.columns[data.columns.str.contains('series_')])
data = data.drop(columns=data.columns[data.columns.str.contains('td_')])
data = data.drop(index=data.index[data.play_type_nfl != 'PASS'])
data = data.drop(columns=data.columns[data.columns.str.contains('penalty')])
data = data.drop(columns=['cp', 'cpoe', 'ep', 'pass_attempt', 'pass_oe', 'sp', 'touchback', 'weather'])
#data = data.drop(columns=data.columns[data.columns.str.contains('end_')])
#data = data.drop(columns=['cp', 'cpoe'])
#data.columns[data.columns.str.contains('end_')]

In [5]:

data.shape

Out[5]:

(18983, 95)

In [6]:

data.columns

Out[6]:

Index(['aborted_play', 'air_yards', 'complete_pass', 'defteam',
       'defteam_score', 'defteam_timeouts_remaining', 'desc', 'div_game',
       'down', 'first_down', 'game_date', 'game_half', 'game_id',
       'game_seconds_remaining', 'goal_to_go', 'half_seconds_remaining', 'id',
       'incomplete_pass', 'interception', 'interception_player_id',
       'interception_player_name', 'location', 'name', 'nfl_api_id',
       'no_huddle', 'old_game_id', 'order_sequence', 'out_of_bounds', 'pass',
       'pass_length', 'pass_location', 'pass_touchdown', 'passer', 'passer_id',
       'passer_player_id', 'passer_player_name', 'passing_yards', 'play',
       'play_clock', 'play_deleted', 'play_id', 'play_type', 'play_type_nfl',
       'posteam', 'posteam_score', 'posteam_timeouts_remaining',
       'posteam_type', 'qb_dropback', 'qb_kneel', 'qb_scramble', 'qb_spike',
       'qtr', 'quarter_end', 'quarter_seconds_remaining', 'receiver',
       'receiver_id', 'receiver_player_id', 'receiver_player_name',
       'receiving_yards', 'replay_or_challenge', 'replay_or_challenge_result',
       'result', 'roof', 'sack', 'safety', 'score_differential', 'season',
       'season_type', 'series', 'shotgun', 'side_of_field', 'special',
       'special_teams_play', 'spread_line', 'st_play_type', 'start_time',
       'success', 'surface', 'temp', 'time', 'time_of_day', 'timeout',
       'timeout_team', 'total', 'total_line', 'touchdown', 'week', 'wind',
       'xpass', 'yardline_100', 'yards_after_catch', 'yards_gained', 'ydsnet',
       'ydstogo', 'yrdln'],
      dtype='object')

Seaborn Violin Plot¶

In [7]:

# https://stackoverflow.com/questions/40372030/pandas-round-to-the-nearest-n
def custom_round(x, base=5):
  return (x // base) * base

In [8]:

def plot_completions_violin(week='ALL', team='ALL'):
  if (team == 'ALL') and (week == 'ALL'):
    plot_subdata = data
  elif (team == 'ALL'):
    plot_subdata = data[(data.week == int(week))]
  elif (week == 'ALL'):
    plot_subdata = data[(data.posteam == team)]
  else:
    plot_subdata = data[(data.week == int(week)) & (data.posteam == team)]

  completions = int(sum(plot_subdata.complete_pass))
  total_passes = plot_subdata.shape[0]
  
  if plot_subdata.shape[0] == 0:
    return None

  plt.figure(figsize=(15, 5))
  
  ax = sns.violinplot(data=plot_subdata,
                      x='air_yards', y='pass_location', order=['left', 'middle', 'right'],
                      hue='complete_pass', split=True, hue_order=[1.0, 0.0],
                      inner='quartile',
                      orient='h',
                      palette={1.0: '#1479FC', 0.0: '0.85'},
                      linewidth=0.5)
  
  ax.xaxis.set_major_locator(plt.MultipleLocator(10))
  ax.xaxis.set_minor_locator(plt.MultipleLocator(5))
  ax.grid(b=True, which='minor', axis='x', linewidth=0.5)

  sns.set_theme(rc={
      'axes.facecolor': '#196F0C',
      'figure.facecolor': '#196F0C',

      'axes.labelcolor': 'white',
      'axes.titlecolor': 'white',
      'axes.titlesize': 20,

      'xtick.color': 'white',
      'ytick.color': 'white',
  })
  plt.setp(ax.collections, edgecolor='black')

  plt.xlabel('Air Yards (from LOS)')
  plt.ylabel('Pass Location')
  plt.title(f'Passing success ({completions}/{total_passes})')
  plt.legend(title = 'Completions')
  
  plt.show()

In [9]:

weeks = data.week.unique()
teams = data.posteam.unique()
teams.sort()

weeks = np.append(weeks, 'ALL')
teams = np.append(teams, 'ALL')

In [14]:

@interact
def f(week=weeks, team=teams):
    return plot_completions_violin(week, team)

interactive(children=(Dropdown(description='week', options=('1', '2', '3', '4', '5', '6', '7', '8', '9', '10',…

Matplotlib Plot¶

In [51]:

#one_game = data[(data.week == 12) & (data.posteam == 'KC')].copy()
def prepare_passing_accuracy_data(one_game):
    one_game['air_yards_rounded'] = custom_round(one_game.air_yards)
    one_game = one_game.groupby(by=['air_yards_rounded', 'pass_location'], as_index=False).agg(
        {
            'complete_pass': ['sum', 'count']     
        }
    )

    one_game.columns = ['air_yards_rounded', 'location', 'completions', 'passes']
    one_game['location_int'] = one_game.location.replace({'left': 2, 'middle': 1, 'right': 0})
    one_game['location'] = one_game.location.str.capitalize()
    one_game['accuracy'] = one_game.completions / one_game.passes
    one_game['text'] = one_game.completions.astype('int64').astype(str) + '/' + one_game.passes.astype(str)
    one_game = one_game.sort_values(by=['location_int'])
    return one_game

In [28]:

def normalize_passes_size(passes):
    max_passes = passes.max()
    if max_passes > 1000:
        min_range, max_range = (1000, 7500)
    elif max_passes > 50:
        min_range, max_range = (1000, 6000)
    else:
        min_range, max_range = (500,  5000)

    return (max_range - min_range) * (passes - passes.min()) / (passes.max() - passes.min()) + min_range    

In [31]:

normalize_passes_size(pd.Series([10000,200, 40, 20]))

Out[31]:

0    7500.000000
1    1117.234469
2    1013.026052
3    1000.000000
dtype: float64

In [33]:

def plot_completions_matplotlib(week='ALL', team='ALL'):
    if (team == 'ALL') and (week == 'ALL'):
        one_game = data
    elif (team == 'ALL'):
        one_game = data[(data.week == int(week))]
    elif (week == 'ALL'):
        one_game = data[(data.posteam == team)]
    else:
        one_game = data[(data.week == int(week)) & (data.posteam == team)]
    
    one_game = prepare_passing_accuracy_data(one_game)
    max_passes = one_game.passes.max()
    total_completions = int(sum(one_game.completions))
    total_passes = int(sum(one_game.passes))

    fig, ax = plt.subplots(1,1,figsize=(20,10))

    # sns.set_theme(rc={
    #     'axes.facecolor': '#196F0C',
    #     'figure.facecolor': '#19300C',

    #     'axes.labelcolor': 'white',
    #     'axes.titlecolor': 'white',
        
    #     'xtick.color': 'white',
    #     'ytick.color': 'white',
    # })

    plt.xlim(-20, 70)
    plt.ylim(-1, 3)
    plt.xticks(fontsize=20, color='white')
    plt.yticks(fontsize=20, color='white')
    
    ax.xaxis.set_major_locator(plt.MultipleLocator(10))
    ax.xaxis.set_minor_locator(plt.MultipleLocator(5))
    ax.grid(b=True, which='both', axis='x', linewidth=2.5)
    #ax.grid(b=True, which='minor', axis='x', linewidth=1)
    ax.grid(b=False, which='both', axis='y')
    ax.set_axisbelow(True)

    # if max_passes > 50:
    #     one_game['passes_size'] = ((6000 - 500) * one_game.passes / max_passes) + 500
    # else:
    #     one_game['passes_size'] = ((5000 - 500) * one_game.passes / max_passes) + 500
    one_game['passes_size'] = normalize_passes_size(one_game.passes)

    plt.scatter(x=one_game.air_yards_rounded,
                y=one_game.location, 
                s=one_game.passes_size,
                c=one_game.accuracy,
                cmap='Oranges',
                edgecolors='black')

    for x, y, text in zip(one_game.air_yards_rounded, one_game.location, one_game.text):
        plt.text(x=x, y=y, s=text, 
                ha = 'center', va = 'center',
                fontsize=15)

    fig.set_facecolor('#196F0C')
    ax.set_facecolor('#196F0C')
    #ax.set_labelcolor('white')
    #ax.set_titlecolor('white')

    plt.xlabel('Air yards (from LOS)', fontsize=15, color='white')
    plt.ylabel('')
    plt.title(f'Pass success for Week={week} and Team={team} ({total_completions}/{total_passes})', fontsize=30, color='white')

    plt.show()
    plt.rcParams.update(plt.rcParamsDefault)

In [34]:

#plot_completions_matplotlib(5, 'KC')
@interact
def f(week=weeks, team=teams):
    return plot_completions_matplotlib(week, team)

interactive(children=(Dropdown(description='week', options=('1', '2', '3', '4', '5', '6', '7', '8', '9', '10',…

Altair Plot¶

In [118]:

team = 'KC'
week = 'ALL'
if (team == 'ALL') and (week == 'ALL'):
    one_game = data
elif (team == 'ALL'):
    one_game = data[(data.week == int(week))]
elif (week == 'ALL'):
    one_game = data[(data.posteam == team)]
else:
    one_game = data[(data.week == int(week)) & (data.posteam == team)]

one_game = prepare_passing_accuracy_data(one_game)
total_completions = int(sum(one_game.completions))
total_passes = int(sum(one_game.passes))

points = alt.Chart(one_game).mark_circle().encode(
    alt.X('air_yards_rounded', title='Air Yards from Line of Scrimmage', 
          axis=alt.Axis(values=list(range(-20,70,5))),
          scale=alt.Scale(domain=[-20, 70])),
    alt.Y('location', title=None),
    alt.Size('passes', legend=None),
    alt.Color('accuracy', scale=alt.Scale(scheme='lightgreyred'), legend=alt.Legend(title='Accuracy')),
)

text = alt.Chart(one_game).mark_text(
    align='center',
    baseline='middle',
    fontSize=10,
    color='black'
).encode(
    x='air_yards_rounded',
    y='location',
    text='text'
)

alt.layer(
    points, 
    text
).properties(
    title=f'Pass success for Week={week} and Team={team} ({total_completions}/{total_passes})',
    #subtitle='here',
    width=1000,
    height=500
).configure(
    background='#196F0C'
).configure_axis(
    labelColor='white',
    labelFontSize=20,
    titleColor='white',
    titleFontSize=30,
).configure_scale(
    maxSize=6000
).configure_title(
    color='white',
    fontSize=30
).configure_view(
    fill='#19500C'
)#.interactive()

Out[118]:

Save as SVG Save as PNG View Source View Compiled Vega Open in Vega Editor

In [205]:

season = data.copy()
season['air_yards_rounded'] = custom_round(season.air_yards)
season = season.groupby(by=['posteam', 'air_yards_rounded', 'pass_location'], as_index=False).agg(
    {
        'complete_pass': ['sum', 'count']     
    }
)

season.columns = ['team', 'air_yards_rounded', 'location', 'completions', 'passes']
season['location_int'] = season.location.replace({'left': 2, 'middle': 1, 'right': 0})
season['location'] = season.location.str.capitalize()
season['accuracy'] = season.completions / season.passes
season['text'] = season.completions.astype('int64').astype(str) + '/' + season.passes.astype(str)
season = season.sort_values(by=['location_int'])
season = season.reset_index()

season_2020_url = 'season_2020.json'
season.to_json(season_2020_url, orient='records')
season.shape

Out[205]:

(1093, 9)

In [208]:

input_dropdown = alt.binding_select(options=np.sort(season.team.unique()), name='Team: ')
selection = alt.selection_single(fields=['team'], bind=input_dropdown, init={'team': 'NE'})
github_season_2020_url = 'https://raw.githubusercontent.com/javendano585/NFL_Data/main/season_2020.json'

points = alt.Chart(github_season_2020_url).mark_circle().encode(
    alt.X('air_yards_rounded:Q', title='Air Yards from Line of Scrimmage', 
          axis=alt.Axis(values=list(range(-20,70,5))),
          scale=alt.Scale(domain=[-20, 70])),
    alt.Y('location:N', title=None),
    alt.Size('passes:Q', legend=None),
    alt.Color('accuracy:Q', scale=alt.Scale(scheme='lightgreyred'), legend=alt.Legend(title='Accuracy')),
    tooltip='text:N'
)

text = alt.Chart(github_season_2020_url).mark_text(
    align='center',
    baseline='middle',
    fontSize=10,
    color='black'
).encode(
    x='air_yards_rounded:Q',
    y='location:N',
    text='text:N'
)

passing_chart = alt.layer(
    points, 
    text
).properties(
    title={
      "text": 'Pass success for 2020 Season', 
      #"subtitle": f'Accuracy = {(total_completions / total_passes):.3f}% ({total_completions}/{total_passes})'
    },
    width=1000,
    height=500
).configure(
    background='#196F0C'
).configure_axis(
    labelColor='white',
    labelFontSize=20,
    titleColor='white',
    titleFontSize=30,
).configure_scale(
    maxSize=6000
).configure_title(
    anchor='start',
    color='white',
    fontSize=30,
    subtitleFontSize=15,
    subtitleColor='white'
).configure_view(
    fill='#19500C'
).add_selection(
    selection
).transform_filter(
    selection
)#.interactive()

passing_chart.save('Passing_2020.html')
passing_chart

Out[208]:

Save as SVG Save as PNG View Source View Compiled Vega Open in Vega Editor