Total Points Model

total-points-model is a Python library with useful functions and notebooks for tuning, training and evaluating an XGBoost model.

The target of this model is the Total Points Scored in an AFL match.

Installation

Use the package manager pip to install total-points-model.

pip install total-points-model

Usage

See notebooks folder for more comprehensive examples of usage.

Data Processing

from total_points_model.domain.preprocessing.data_preprocessor import DataPreprocessor
from total_points_model.domain.contracts.mappings import Mappings
from total_points_model.domain.contracts.rolling_columns import RollingColumns

preprocessor = DataPreprocessor(Mappings=Mappings, rolling_dict=RollingColumns.rolling_dict)

preprocessor.fit(training_data)

training_data_preproc = preprocessor.transform(training_data)
test_data_preproc = preprocessor.transform(test_data)

Hyperparameter Tuning

from total_points_model.domain.modelling.hyperparameter_tuning import XGBYearHyperparameterTuner
from total_points_model.domain.modelling.optuna_xgb_param_grid import OptunaXGBParamGrid

xgb_tuner = XGBYearHyperparameterTuner(X_train_preproc, y_train, optuna_grid=OptunaXGBParamGrid, monotonicity_constraints={})
xgb_tuner.tune_hyperparameters()
params = xgb_tuner.get_best_params()

Model Building

from total_points_model.domain.modelling.supermodel import SuperXGBRegressor


super_xgb = SuperXGBRegressor(X_train = X_train_preproc, 
                              y_train = y_train, 
                              X_test = X_test_preproc, 
                              y_test = y_test, 
                              params = params)

super_xgb.fit()

train_preds = super_xgb.predict(X_train_preproc.drop(columns=["Match_ID"]))
test_preds = super_xgb.predict(X_test_preproc.drop(columns=["Match_ID"]))

Model Evaluation

from total_points_model.domain.modelling.model_evaluation import XGBRegressorEvaluator
from total_points_model.domain.contracts.modelling_data_contract import ModellingDataContract

train_xgb_evals = XGBRegressorEvaluator(model = xgb_model,
                                       data = train_evaluation_data,
                                       actual_name = ModellingDataContract.RESPONSE,
                                       expected_name = "xgb_preds",
                                       compare_name='old_preds'
                                       )
train_xgb_evals.plot_distribution()

test_xgb_evals = XGBRegressorEvaluator(model = xgb_model,
                                       data = test_evaluation_data,
                                       actual_name = ModellingDataContract.RESPONSE,
                                       expected_name = "xgb_preds",
                                       compare_name = "old_preds"
                                       )
test_xgb_evals.plot_distribution()

Credits

Data sourced using a private R package. Credits to dgt23.

CONTRIBUTING

I am no longer updating this package as I am exploring other AFL related analysis and packages.

License

MIT

ciaran-grant/afl-total-points-model