Skip to main content

SSAT: Statistical Sports Analysis Toolkit

Project description

SSAT: Statistical Sports Analysis Toolkit

Python 3.10+ License: MIT Code style: ruff PyPI

SSAT is a comprehensive Python package for statistical sports analysis, providing both frequentist and Bayesian statistical models for analyzing and predicting sports match outcomes. The package is particularly focused on handball but can be adapted for other sports.

🚀 Key Features

Statistical Models

  • Frequentist Models: Bradley-Terry, GSSD, TOOR, ZSD, PRP, Poisson
  • Bayesian Models: Poisson, Negative Binomial, Skellam variants with MCMC sampling
  • Model Comparison: Built-in tools for comparing predictions across different approaches

Analysis Capabilities

  • Team Ratings: Detailed offensive/defensive capabilities analysis
  • Match Prediction: Win/Draw/Loss probabilities with uncertainty quantification
  • Performance Evaluation: Comprehensive model benchmarking and validation
  • Visualization: Rich plotting utilities for model diagnostics and team analysis

Data Integration

  • Sample Data: Included handball datasets for immediate experimentation
  • Flexible Input: Support for various data formats and structures
  • Extensible: Easy integration with external data sources

✅ To-Do

  • Streamlined fit and predic for both types of models

  • Write documentation to show features like adding weights

📦 Installation

Installation

pip install ssat

cmdStan Intallation

To use the Bayesian models you will need to install cmdStan as described in the cmdStan Installation Guide.

🏃‍♂️ Quick Start

Frequentist Models Example

import pandas as pd
from ssat.data import get_sample_handball_match_data
from ssat.frequentist import GSSD, BradleyTerry

# Load sample data
df = get_sample_handball_match_data()
league = "Starligue"
season = 2024
match_df = df.loc[(df["league"] == league) & (df["season"] == season)]

# Prepare data
X = match_df[["home_team", "away_team"]]
y = match_df["home_goals"] - match_df["away_goals"]  # spread
Z = match_df[["home_goals", "away_goals"]]

# Train-test split
train_size = int(len(match_df) * 0.8)
X_train, X_test = X[:train_size], X[train_size:]
y_train, y_test = y[:train_size], y[train_size:]
Z_train, Z_test = Z[:train_size], Z[train_size:]

# Fit models
bt_model = BradleyTerry()
bt_model.fit(X_train, y_train, Z_train)

gssd_model = GSSD()
gssd_model.fit(X_train, y_train, Z_train)

# Make predictions
test_fixtures = X_test.apply(lambda x: f"{x.home_team}-{x.away_team}", axis=1)

bt_probas = bt_model.predict_proba(X_test, point_spread=0, include_draw=True)
gssd_probas = gssd_model.predict_proba(X_test, point_spread=0, include_draw=True)

bt_probas_df = pd.DataFrame(
    bt_probas, columns=["Home", "Draw", "Away"], index=test_fixtures
)
gssd_probas_df = pd.DataFrame(
    gssd_probas, columns=["Home", "Draw", "Away"], index=test_fixtures
)
print(bt_probas_df.head())
print(gssd_probas_df.head())

# Get team ratings
bt_team_ratings = bt_model.get_team_ratings()
print(bt_team_ratings.head())

gssd_team_ratings = gssd_model.get_team_ratings()
print(gssd_team_ratings.head())

Bayesian Models Example

import pandas as pd
from ssat.bayesian import Poisson, Skellam
from ssat.data import get_sample_handball_match_data

# Load sample data
df = get_sample_handball_match_data()
league = "Starligue"
season = 2024
match_df = df.loc[(df["league"] == league) & (df["season"] == season)]

# Prepare data
X = match_df[["home_team", "away_team", "home_goals", "away_goals"]]
X = X.assign(goal_diff=X["home_goals"] - X["away_goals"])

# Train-test split
train_size = int(len(match_df) * 0.8)
X_train, X_test = X[:train_size], X[train_size:]

# Fit Bayesian models
poisson_model = Poisson()
poisson_model.fit(X_train, seed=42)

skellam_model = Skellam()
skellam_model.fit(X_train[["home_team", "away_team", "goal_diff"]], seed=42)

# Visualize model diagnostics
poisson_model.plot_trace()
poisson_model.plot_team_stats()

skellam_model.plot_trace()
skellam_model.plot_team_stats()

# Make predictions on new matches
test_fixtures = X_test.apply(lambda x: f"{x.home_team}-{x.away_team}", axis=1)

poisson_preds = poisson_model.predict(X_test)
poisson_probas = poisson_model.predict_proba(X_test)
poisson_probas.index = test_fixtures

skellam_preds = skellam_model.predict(X_test)
skellam_probas = skellam_model.predict_proba(X_test)
skellam_probas.index = test_fixtures

# Print results - notice how the Skellam assign a higher probability to draws
print(poisson_probas.head())
print(skellam_probas.head())

📊 Model Overview

Frequentist Models

Model Description
Bradley-Terry Paired comparison with logistic regression
GSSD Linear regression with offensive/defensive stats
TOOR Team offense-offense rating
ZSD Zero-score distribution modeling
PRP Possession-based rating process
Poisson Goal-scoring as Poisson process

Bayesian Models

Model Description
Poisson Bayesian goal-scoring with MCMC
NegBinom Overdispersed goal modeling
Skellam Direct goal difference modeling
SkellamZero Zero-inflated Skellam
Weighted variants Time-weighted model fitting

📈 Example Notebooks

The repository contains comprehensive example notebooks:

Both examples use real handball data and demonstrate:

  • Proper train-test splitting
  • Model performance evaluation
  • Prediction comparison and visualization
  • Team strength analysis

🔧 Other Usage

Model Benchmarking

import numpy as np
import pandas as pd
from ssat.data import get_sample_handball_match_data
from ssat.frequentist import BradleyTerry, GSSD

# Load sample data
df = get_sample_handball_match_data()
league = "Starligue"
season = 2024
match_df = df.loc[(df["league"] == league) & (df["season"] == season)]

# Prepare data
X = match_df[["home_team", "away_team"]]
y = match_df["home_goals"] - match_df["away_goals"]  # spread
Z = match_df[["home_goals", "away_goals"]]

# Train-test split
train_size = int(len(match_df) * 0.8)
X_train, X_test = X[:train_size], X[train_size:]
y_train, y_test = y[:train_size], y[train_size:]
Z_train, Z_test = Z[:train_size], Z[train_size:]

# Compare multiple models
models = [BradleyTerry(), GSSD()]
results = {}

for model in models:
    model.fit(X_train, y_train, Z_train)
    preds = model.predict(X_test)
    results[model.NAME] = np.mean(np.abs(preds - y_test))

print("Model Performance (MAE):")
for model_name, mae in results.items():
    print(f"{model_name}: {mae:.3f}")

Custom Team Analysis

import pandas as pd
from ssat.data import get_sample_handball_match_data
from ssat.frequentist import BradleyTerry, GSSD

# Load sample data
df = get_sample_handball_match_data()
league = "Starligue"
season = 2024
match_df = df.loc[(df["league"] == league) & (df["season"] == season)]


# Prepare data
X = match_df[["home_team", "away_team"]]
y = match_df["home_goals"] - match_df["away_goals"]  # spread
Z = match_df[["home_goals", "away_goals"]]

# Train-test split
train_size = int(len(match_df) * 0.8)
X_train, X_test = X[:train_size], X[train_size:]
y_train, y_test = y[:train_size], y[train_size:]
Z_train, Z_test = Z[:train_size], Z[train_size:]

# Fit Model
gssd_model = GSSD()
gssd_model.fit(X_train, y_train, Z_train)

# Detailed team strength analysis
team_stats = gssd_model.get_team_ratings()
print("Team Offensive/Defensive Breakdown:")
print(team_stats[['pfh', 'pah', 'pfa', 'paa']].head())

# Model coefficients
coeffs = team_stats.loc['Coefficients']
print(f"Home offense coefficient: {coeffs['pfh']:.3f}")
print(f"Home defense coefficient: {coeffs['pah']:.3f}")

📊 Data Format

SSAT expects data in the following format:

# Required columns for match data
match_data = pd.DataFrame({
    'home_team': ['Team A', 'Team B', ...],
    'away_team': ['Team B', 'Team C', ...],
    'home_goals': [25, 30, ...],
    'away_goals': [23, 28, ...],
})

🛠️ Development

Setup Development Environment

git clone https://github.com/bjrnsa/ssat.git
cd ssat
# Create and activate your virtual environment
pip install -e .

Run Examples or check out the rendered notebooks in the ssat/notebooks folder

# Frequentist models example
python ssat/notebooks/frequentist_example.py

# Bayesian models example
python ssat/notebooks/bayesian_example.py

📝 Dependencies

Core Dependencies

  • arviz: Bayesian model diagnostics
  • cmdstanpy: Stan interface for MCMC sampling
  • matplotlib: Plotting and visualization
  • numpy: Numerical computing
  • pandas: Data manipulation
  • pyarrow: Efficient data storage
  • scipy: Statistical functions
  • seaborn: Statistical visualization

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

📚 Citation

If you use SSAT in your research, please cite:

@software{ssat2025,
  author = {Aagaard, Bjørn},
  title = {SSAT: Statistical Sports Analysis Toolkit},
  version = {0.0.3},
  year = {2025},
  publisher = {GitHub},
  url = {https://github.com/bjrnsa/ssat}
}

🙏 Acknowledgments

  • Statistical modeling concepts from Andrew Mack's "Statistical Sports Models in Excel"
  • The Stan development team for excellent MCMC tools
  • The scientific Python ecosystem contributors

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ssat-0.0.4.tar.gz (466.6 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

ssat-0.0.4-py3-none-any.whl (377.5 kB view details)

Uploaded Python 3

File details

Details for the file ssat-0.0.4.tar.gz.

File metadata

  • Download URL: ssat-0.0.4.tar.gz
  • Upload date:
  • Size: 466.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.7.13

File hashes

Hashes for ssat-0.0.4.tar.gz
Algorithm Hash digest
SHA256 69112e75b4d7d63e27b1cc0cf20a8d4e410c52737381670e2a20d8b93185fc70
MD5 59bd3b69a3a595ee61694182151a87cf
BLAKE2b-256 d78cc6c5f99745da8515342d5dc732673fd3b908387e7da42d75a6fd84297eb8

See more details on using hashes here.

File details

Details for the file ssat-0.0.4-py3-none-any.whl.

File metadata

  • Download URL: ssat-0.0.4-py3-none-any.whl
  • Upload date:
  • Size: 377.5 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.7.13

File hashes

Hashes for ssat-0.0.4-py3-none-any.whl
Algorithm Hash digest
SHA256 fc63775338546c2556c2828085376d4c08dcee7a9e515e81e5af973656c630b9
MD5 3f6ce4bbb393c8476f859a130a5df013
BLAKE2b-256 1b6b91fb74b2ac2b615c377d85acdf77faf76633d2c154291eecfae12aa40a45

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page