Rolling xG tutorial
Intro
Use the chickenstats library to scrape play-by-play data and plot rolling average xGF and xGA.
Parts of this tutorial are optional and will be clearly marked as such. For help, or any questions, please don't hesitate to reach out to chicken@chickenandstats.com or @chickenandstats.com on Blue Sky.
Housekeeping
Import dependencies
Import the dependencies we'll need for the guide
import datetime as dt
from pathlib import Path
import matplotlib.patches as patches
import matplotlib.patheffects as mpe
import matplotlib.pyplot as plt
import matplotlib.ticker as mtick
import matplotlib.ticker as ticker
import numpy as np
import pandas as pd
import seaborn as sns
import chickenstats.utilities
from chickenstats.chicken_nhl import Scraper, Season
from chickenstats.chicken_nhl.info import NHL_COLORS
from chickenstats.chicken_nhl.helpers import charts_directory
Pandas options
Sets different pandas options. This cell is optional
Folder structure
Chickenstats matplotlib style
chickenstats.utilities includes a custom style package - this activates it. This cell is also optional
Scrape data
Schedule, standings, and team names
Scrape the schedule and standings using the Season object. Then, create a list of game IDs to scrape
condition = schedule.game_state == "OFF"
game_ids = schedule.loc[condition].game_id.tolist()
latest_date = schedule.loc[condition].game_date.max()
Play-by-play
Scrape the play-by-play data for the chosen game ID. First instantiate the Scraper object,
then call the play_by_play attribute
Stats
Aggregate statistics
scraper.prep_stats(level="season", disable_progress_bar=True)
stats = scraper.stats.reset_index(drop=True)
scraper.prep_lines(level="season", disable_progress_bar=True)
lines = scraper.lines.reset_index(drop=True)
scraper.prep_team_stats(level="game", disable_progress_bar=True)
team_stats = scraper.team_stats.reset_index(drop=True)
Helper function
Generates the rolling average figures for the specific team
def get_xg_rolling_data(data: pd.DataFrame, season: str, session: str, team: str, strengths: list, window: int = 10):
"""This function returns rolling average xG figures for a specific team.
Parameters:
data (pd.DataFrame):
Pandas dataframe of team stats aggregated from play-by-play data scraped with the chickenstats package
season (str):
8-digit season code (start year + end year) as a string
session (str):
"R" or "P," determines regular season or playoffs
team (str):
Three-digit team code to filter the data
strengths (list):
Strength states, e.g., 5v5, 5v4, to include in the grouping, as a list
window (int):
Number of games for the rolling average calculation, default is 10
"""
df = data.copy()
conds = [df.season == season, df.session == session, df.team == team]
game_num = df[np.logical_and.reduce(conds)].game_id.unique()
num_map = {x: idx + 1 for idx, x in enumerate(game_num)}
conds = [df.season == season, df.session == session, df.team == team, df.strength_state.isin(strengths)]
df = df[np.logical_and.reduce(conds)].copy()
df["game_num"] = df.game_id.map(num_map)
for_list = ["cf_p60", "ff_p60", "hdff_p60", "sf_p60", "hdsf_p60", "gf_p60", "hdgf_p60", "xgf_p60"]
against_list = ["ca_p60", "fa_p60", "hdfa_p60", "sa_p60", "hdsa_p60", "ga_p60", "hdga_p60", "xga_p60"]
stats_dict = dict(zip(for_list, against_list, strict=False))
for f, a in stats_dict.items():
df[f"rolling_{f}"] = df[f].rolling(window=window, min_periods=0).mean()
df[f"rolling_{a}"] = df[a].rolling(window=window, min_periods=0).mean()
df[f"rolling_{f}_diff"] = df[f"rolling_{f}"] - df[f"rolling_{a}"]
return df
Plotting data
Whole NHL
Plot each individual NHL team as its own subplot
# Setting overall figures
fig, axes = plt.subplots(nrows=8, ncols=4, dpi=650, figsize=(12, 18))
fig.tight_layout(pad=1.5)
axes = axes.reshape(-1)
# Getting the teams and standings data to iterate through
teams = standings.team.unique().tolist()
team_names = dict(zip(standings.team, standings.team_name, strict=False))
# Iterating through the standings data
for idx, row in standings.iterrows():
# Setting the team
team = row.team
# Setting the axis
ax = axes[idx]
# Setting uniform y limit
ax.set_ylim(0.75, 7.0)
ax.set_xlim(0, standings.games_played.max() + 2)
# Getting df for plotting
df = get_xg_rolling_data(team_stats, year, session, team, strengths, window=3)
# Getting the Y data to plot
Y_for = df.rolling_xgf_p60.copy().reset_index(drop=True)
Y_ag = df.rolling_xga_p60.copy().reset_index(drop=True)
# Getting the X data to plot
X = pd.Series(range(1, max(df.game_num) + 1))
# Setting colors
colors = NHL_COLORS[team]
for_c = colors["GOAL"]
ag_c = colors["SHOT"]
# Setting path effects for xGF line
pe_ec = ag_c if for_c == "#FFFFFF" else "white"
pe_for = [
mpe.Stroke(linewidth=3.25, foreground=for_c),
mpe.Stroke(foreground=pe_ec, alpha=1, linewidth=4),
mpe.Normal(),
]
# Plotting xGF
sns.lineplot(x=X, y=Y_for, color=for_c, ax=ax, zorder=3, path_effects=pe_for)
# Setting path effects for xGA line
pe_ec = for_c if ag_c == "#FFFFFF" else "white"
pe_ag = [
mpe.Stroke(linewidth=3.25, foreground=ag_c),
mpe.Stroke(foreground=pe_ec, alpha=1, linewidth=4),
mpe.Normal(),
]
# Plotting xGA line
sns.lineplot(x=X, y=Y_ag, color=ag_c, ax=ax, zorder=3, path_effects=pe_ag)
# Changing colors if for color is white
if for_c == "#FFFFFF":
# Filling between lines
ax.fill_between(
X,
Y_ag,
Y_for,
where=Y_for > Y_ag,
interpolate=True,
alpha=0.9,
zorder=2,
facecolor=for_c,
edgecolor=ag_c,
hatch="/////",
lw=1,
)
# Setting path effect for legend
pe_for = [
mpe.Stroke(linewidth=3.25, foreground=for_c),
mpe.Stroke(foreground=ag_c, alpha=1, linewidth=4),
mpe.Normal(),
]
# Setting the legend figures
xgf_fill = patches.Patch(facecolor=for_c, edgecolor=ag_c, hatch="/////", label="+xG DIFFERENTIAL")
xgf_l = patches.Patch(facecolor=for_c, label="xG FOR", edgecolor=ag_c)
else:
# Fill between the lines
ax.fill_between(X, Y_ag, Y_for, where=Y_for > Y_ag, interpolate=True, alpha=0.9, zorder=2, color=for_c)
# Setting the legend figures
xgf_fill = patches.Patch(facecolor=for_c, edgecolor=for_c, label="+xG DIFFERENTIAL")
xgf_l = patches.Patch(facecolor=for_c, label="xG FOR", edgecolor=for_c)
if ag_c == "#FFFFFF":
# Fill between the lines
ax.fill_between(
X,
Y_ag,
Y_for,
where=Y_ag >= Y_for,
interpolate=True,
alpha=0.9,
zorder=2,
edgecolor=for_c,
facecolor=ag_c,
hatch="////",
lw=1,
)
# Setting the legend figures
xga_fill = patches.Patch(facecolor=ag_c, edgecolor=for_c, hatch="/////", label="-xG DIFFERENTIAL")
xga_l = patches.Patch(facecolor=ag_c, label="xG AGAINST", edgecolor=for_c)
else:
# Fill between the lines
ax.fill_between(X, Y_ag, Y_for, where=Y_ag >= Y_for, interpolate=True, alpha=0.9, zorder=2, color=ag_c)
# Setting the legend figures
xga_fill = patches.Patch(facecolor=ag_c, edgecolor=ag_c, label="-xG DIFFERENTIAL")
xga_l = patches.Patch(facecolor=ag_c, label="xG AGAINST", edgecolor=ag_c)
# Setting y-axis major locator
ax.yaxis.set_major_locator(mtick.MultipleLocator(1))
# Summary stats
stats = df
gf = stats.gf.sum()
xgf = stats.xgf.sum()
ga = stats.ga.sum()
xga = stats.xga.sum()
# Subtitle text
textstr = f"{gf} GF - {ga} GA | {round(xgf, 2)} xGF - {round(xga, 2)} xGA"
props = dict(boxstyle="round", facecolor="white", alpha=0.9, lw=0.8, ec="white")
ax.text(0.5, 0.99, textstr, transform=ax.transAxes, fontsize=7, ha="center", va="center", bbox=props)
# Legend elements
legend_elements = [xgf_l, xga_l, xgf_fill, xga_fill]
ax.legend(
handles=legend_elements, loc="upper center", bbox_to_anchor=(0.5, 0.96), fontsize=5, ncol=2, borderpad=0.55
)
# Setting x and y axes labels
x_labels = [28, 29, 30, 31]
if idx in x_labels:
ax.axes.set_xlabel("Game number", fontsize=8)
else:
ax.axes.set_xlabel("")
y_labels = [0, 4, 8, 12, 16, 20, 24, 28]
if idx in y_labels:
ax.axes.set_ylabel(" 3-game rolling avg. xG / 60", fontsize=8)
else:
ax.axes.set_ylabel("")
# Setting tick params font size
ax.tick_params(axis="both", which="major", labelsize=8)
# Setting the ax title
ax_title = f"{row.team_name} | {row.points} points | {row.wins} - {row.losses} - {row.ot_losses}"
ax.set_title(ax_title, fontsize=8, x=-0.085, y=1.03, horizontalalignment="left")
# Figure suptitle and subtitle
fig_suptitle = "3-game rolling average 5v5 chances created vs. allowed (per 60 minutes)"
fig.suptitle(fig_suptitle, x=0.01, y=1.029, fontsize=11, fontweight="bold", horizontalalignment="left")
subtitle = f"5v5 team xGF / 60 and xGA / 60 | 2024-25 season, as of {latest_date}"
fig.text(s=subtitle, x=0.01, y=1.0115, fontsize=10, horizontalalignment="left")
# Attribution
attribution = "Data & xG model @chickenandstats.com | Viz @chickenandstats.com"
fig.text(s=attribution, x=0.99, y=-0.01, fontsize=8, horizontalalignment="right", style="italic")
save_path = Path("./charts/nhl_rolling_xg.png")
fig.savefig(save_path, transparent=False, bbox_inches="tight")
Individual team
Plot light and dark mode for an individual team
Light mode
# Getting the teams and standings data to iterate through
teams = standings.team.unique().tolist()
team_names = dict(zip(standings.team, standings.team_name, strict=False))
with plt.style.context("chickenstats"):
fig, ax = plt.subplots(dpi=650, figsize=(8, 5))
# Getting df for plotting
df = get_xg_rolling_data(team_stats, year, session, team, strengths, window=3)
# Getting the Y data to plot
Y_for = df.rolling_xgf_p60.copy().reset_index(drop=True)
Y_ag = df.rolling_xga_p60.copy().reset_index(drop=True)
# Getting the X data to plot
X = pd.Series(range(1, max(df.game_num) + 1))
# Setting colors
colors = NHL_COLORS[team]
for_c = colors["GOAL"]
ag_c = colors["SHOT"]
# Setting path effects for xGF line
pe_ec = ag_c if for_c == "#FFFFFF" else "white"
pe_for = [
mpe.Stroke(linewidth=3.25, foreground=for_c),
mpe.Stroke(foreground=pe_ec, alpha=1, linewidth=4),
mpe.Normal(),
]
# Plotting xGF
sns.lineplot(x=X, y=Y_for, color=for_c, ax=ax, zorder=3, path_effects=pe_for)
# Setting path effects for xGA line
pe_ec = for_c if ag_c == "#FFFFFF" else "white"
pe_ag = [
mpe.Stroke(linewidth=3.25, foreground=ag_c),
mpe.Stroke(foreground=pe_ec, alpha=1, linewidth=4),
mpe.Normal(),
]
# Plotting xGA line
sns.lineplot(x=X, y=Y_ag, color=ag_c, ax=ax, zorder=3, path_effects=pe_ag)
# Changing colors if for color is white
if for_c == "#FFFFFF":
# Filling between lines
ax.fill_between(
X,
Y_ag,
Y_for,
where=Y_for > Y_ag,
interpolate=True,
alpha=0.9,
zorder=2,
facecolor=for_c,
edgecolor=ag_c,
hatch="/////",
lw=1,
)
# Setting path effect for legend
pe_for = [
mpe.Stroke(linewidth=3.25, foreground=for_c),
mpe.Stroke(foreground=ag_c, alpha=1, linewidth=4),
mpe.Normal(),
]
# Setting the legend figures
xgf_fill = patches.Patch(facecolor=for_c, edgecolor=ag_c, hatch="/////", label="+xG DIFFERENTIAL")
xgf_l = patches.Patch(facecolor=for_c, label="xG FOR", edgecolor=for_c)
else:
# Fill between the lines
ax.fill_between(X, Y_ag, Y_for, where=Y_for > Y_ag, interpolate=True, alpha=0.9, zorder=2, color=for_c)
# Setting the legend figures
xgf_fill = patches.Patch(facecolor=for_c, edgecolor=for_c, label="+xG DIFFERENTIAL")
xgf_l = patches.Patch(facecolor=for_c, label="xG FOR", edgecolor=for_c)
if ag_c == "#FFFFFF":
# Fill between the lines
ax.fill_between(
X,
Y_ag,
Y_for,
where=Y_ag >= Y_for,
interpolate=True,
alpha=0.9,
zorder=2,
edgecolor=for_c,
facecolor=ag_c,
hatch="////",
lw=1,
)
# Setting the legend figures
xga_fill = patches.Patch(facecolor=ag_c, edgecolor=for_c, hatch="/////", label="-xG DIFFERENTIAL")
xga_l = patches.Patch(facecolor=ag_c, label="xG AGAINST", edgecolor=ag_c)
else:
# Fill between the lines
ax.fill_between(X, Y_ag, Y_for, where=Y_ag >= Y_for, interpolate=True, alpha=0.9, zorder=2, color=ag_c)
# Setting the legend figures
xga_fill = patches.Patch(facecolor=ag_c, edgecolor=ag_c, label="-xG DIFFERENTIAL")
xga_l = patches.Patch(facecolor=ag_c, label="xG AGAINST", edgecolor=ag_c)
# Setting y-axis major locator
ax.yaxis.set_major_locator(mtick.MultipleLocator(1))
# Summary stats
stats = df
gf = stats.gf.sum()
xgf = stats.xgf.sum()
ga = stats.ga.sum()
xga = stats.xga.sum()
# Legend elements
legend_elements = [xgf_l, xga_l, xgf_fill, xga_fill]
ax.legend(
handles=legend_elements, loc="upper left", fontsize=8, ncol=2, borderpad=0.55, framealpha=0, edgecolor="white"
)
ax.axes.set_xlabel("Game number", fontsize=8)
ax.axes.set_ylabel(" 3-game rolling avg. xG / 60", fontsize=8)
ax.xaxis.set_major_locator(ticker.MultipleLocator(5))
ax.xaxis.set_minor_locator(ticker.MultipleLocator(1))
ax.yaxis.set_major_locator(ticker.MultipleLocator(1))
ax.yaxis.set_minor_locator(ticker.MultipleLocator(0.25))
# Setting tick params font size
ax.tick_params(axis="both", which="major", labelsize=8)
standings_team = standings.loc[standings.team == team].iloc[0]
ax_title = f"{standings_team.team_name}"
# Setting the ax title
ax_title = f"{ax_title}"
ax.set_title(ax_title, fontsize=10, x=-0.05, y=1.05, horizontalalignment="left")
subtitle_standings = (
f"{standings_team.points} points ({standings_team.wins} - {standings_team.losses} - {standings_team.ot_losses})"
)
subtitle_goals = f"{gf} GF ({round(xgf, 2)} xGF) - {ga} GA ({round(xga, 2)} xGA)"
ax_subtitle = f"{subtitle_standings} | {subtitle_goals} at 5v5"
ax.text(s=ax_subtitle, fontsize=9, x=-0.05, y=1.025, horizontalalignment="left", transform=ax.transAxes)
attribution = "Data & xG model @chickenandstats.com | Viz @chickenandstats.com"
ax.text(
s=attribution, fontsize=7, x=1, y=-0.15, horizontalalignment="right", transform=ax.transAxes, fontstyle="italic"
)
save_path = Path(f"./charts/{team.lower()}_rolling_xg.png")
fig.savefig(save_path, transparent=False, bbox_inches="tight")
Dark mode
team = "NSH"
year = "20242025"
strengths = ["5v5"]
session = "R"
# Getting the teams and standings data to iterate through
teams = standings.team.unique().tolist()
team_names = dict(zip(standings.team, standings.team_name, strict=False))
with plt.style.context("chickenstats_dark"):
fig, ax = plt.subplots(dpi=650, figsize=(8, 5))
# Getting df for plotting
df = get_xg_rolling_data(team_stats, year, session, team, strengths, window=3)
# Getting the Y data to plot
Y_for = df.rolling_xgf_p60.copy().reset_index(drop=True)
Y_ag = df.rolling_xga_p60.copy().reset_index(drop=True)
# Getting the X data to plot
X = pd.Series(range(1, max(df.game_num) + 1))
# Setting colors
colors = NHL_COLORS[team]
for_c = colors["GOAL"]
ag_c = colors["SHOT"]
# Setting path effects for xGF line
pe_ec = ag_c if for_c == "#FFFFFF" else "white"
pe_for = [
mpe.Stroke(linewidth=3.25, foreground=for_c),
mpe.Stroke(foreground=pe_ec, alpha=1, linewidth=4),
mpe.Normal(),
]
# Plotting xGF
sns.lineplot(x=X, y=Y_for, color=for_c, ax=ax, zorder=3, path_effects=pe_for)
# Setting path effects for xGA line
pe_ec = for_c if ag_c == "#FFFFFF" else "white"
pe_ag = [
mpe.Stroke(linewidth=3.25, foreground=ag_c),
mpe.Stroke(foreground=pe_ec, alpha=1, linewidth=4),
mpe.Normal(),
]
# Plotting xGA line
sns.lineplot(x=X, y=Y_ag, color=ag_c, ax=ax, zorder=3, path_effects=pe_ag)
# Changing colors if for color is white
if for_c == "#FFFFFF":
# Filling between lines
ax.fill_between(
X,
Y_ag,
Y_for,
where=Y_for > Y_ag,
interpolate=True,
alpha=0.9,
zorder=2,
facecolor=for_c,
edgecolor=ag_c,
hatch="/////",
lw=1,
)
# Setting path effect for legend
pe_for = [
mpe.Stroke(linewidth=3.25, foreground=for_c),
mpe.Stroke(foreground=ag_c, alpha=1, linewidth=4),
mpe.Normal(),
]
# Setting the legend figures
xgf_fill = patches.Patch(facecolor=for_c, edgecolor="white", hatch="/////", label="+xG DIFFERENTIAL")
xgf_l = patches.Patch(facecolor=for_c, label="xG FOR", edgecolor="white")
else:
# Fill between the lines
ax.fill_between(X, Y_ag, Y_for, where=Y_for > Y_ag, interpolate=True, alpha=0.9, zorder=2, color=for_c)
# Setting the legend figures
xgf_fill = patches.Patch(facecolor=for_c, edgecolor="white", label="+xG DIFFERENTIAL")
xgf_l = patches.Patch(facecolor=for_c, label="xG FOR", edgecolor="white")
if ag_c == "#FFFFFF":
# Fill between the lines
ax.fill_between(
X,
Y_ag,
Y_for,
where=Y_ag >= Y_for,
interpolate=True,
alpha=0.9,
zorder=2,
edgecolor=for_c,
facecolor=ag_c,
hatch="////",
lw=1,
)
# Setting the legend figures
xga_fill = patches.Patch(facecolor=ag_c, edgecolor="white", hatch="/////", label="-xG DIFFERENTIAL")
xga_l = patches.Patch(facecolor=ag_c, label="xG AGAINST", edgecolor="white")
else:
# Fill between the lines
ax.fill_between(X, Y_ag, Y_for, where=Y_ag >= Y_for, interpolate=True, alpha=0.9, zorder=2, color=ag_c)
# Setting the legend figures
xga_fill = patches.Patch(facecolor=ag_c, edgecolor="white", label="-xG DIFFERENTIAL")
xga_l = patches.Patch(facecolor=ag_c, label="xG AGAINST", edgecolor="white")
# Setting y-axis major locator
ax.yaxis.set_major_locator(mtick.MultipleLocator(1))
# Summary stats
stats = df
gf = stats.gf.sum()
xgf = stats.xgf.sum()
ga = stats.ga.sum()
xga = stats.xga.sum()
# Legend elements
legend_elements = [xgf_l, xga_l, xgf_fill, xga_fill]
ax.legend(
handles=legend_elements, loc="upper left", fontsize=8, ncol=2, borderpad=0.55, framealpha=0, edgecolor="white"
)
ax.axes.set_xlabel("Game number", fontsize=8)
ax.axes.set_ylabel(" 3-game rolling avg. xG / 60", fontsize=8)
ax.xaxis.set_major_locator(ticker.MultipleLocator(5))
ax.xaxis.set_minor_locator(ticker.MultipleLocator(1))
ax.yaxis.set_major_locator(ticker.MultipleLocator(1))
ax.yaxis.set_minor_locator(ticker.MultipleLocator(0.25))
# Setting tick params font size
ax.tick_params(axis="both", which="major", labelsize=8)
standings_team = standings.loc[standings.team == team].iloc[0]
ax_title = f"{standings_team.team_name}"
# Setting the ax title
ax_title = f"{ax_title}"
ax.set_title(ax_title, fontsize=10, x=-0.05, y=1.05, horizontalalignment="left")
subtitle_standings = (
f"{standings_team.points} points ({standings_team.wins} - {standings_team.losses} - {standings_team.ot_losses})"
)
subtitle_goals = f"{gf} GF ({round(xgf, 2)} xGF) - {ga} GA ({round(xga, 2)} xGA)"
ax_subtitle = f"{subtitle_standings} | {subtitle_goals} at 5v5"
ax.text(s=ax_subtitle, fontsize=9, x=-0.05, y=1.025, horizontalalignment="left", transform=ax.transAxes)
attribution = "Data & xG model @chickenandstats.com | Viz @chickenandstats.com"
ax.text(
s=attribution, fontsize=7, x=1, y=-0.15, horizontalalignment="right", transform=ax.transAxes, fontstyle="italic"
)
save_path = Path(f"./charts/{team.lower()}_rolling_xg_dark.png")
fig.savefig(save_path, transparent=False, bbox_inches="tight")
