This repository contains Python bindings to the XGP library. It is a simple wrapper that calls the XGP dynamic shared library and exposes a scikit-learn interface.
Please refer to the Python section of the XGP website.
Installation instructions are available here.
>>> from sklearn import datasets
>>> from sklearn import metrics
>>> from sklearn import model_selection
>>> import xgp
>>> X, y = datasets.load_boston(return_X_y=True)
>>> X_train, X_test, y_train, y_test = model_selection.train_test_split(X, y, random_state=42)
>>> model = xgp.XGPRegressor(
... flavor='boosting',
... loss_metric='mse',
... funcs='add,sub,mul,div',
... n_individuals=50,
... n_generations=20,
... parsimony_coefficient=0.01,
... n_rounds=8,
... random_state=42,
... )
>>> model = model.fit(X_train, y_train, eval_set=(X_test, y_test), verbose=True)
>>> metrics.mean_squared_error(y_train, model.predict(X_train)) # doctest: +ELLIPSIS
17.794685...
>>> metrics.mean_squared_error(y_test, model.predict(X_test)) # doctest: +ELLIPSIS
17.337693...This will also produce the following output in the shell:
00:00:00 -- train mse: 42.06567 -- val mse: 33.80606 -- round 1
00:00:00 -- train mse: 24.20662 -- val mse: 22.73832 -- round 2
00:00:00 -- train mse: 22.06328 -- val mse: 18.90887 -- round 3
00:00:00 -- train mse: 20.25549 -- val mse: 18.45531 -- round 4
00:00:00 -- train mse: 18.86693 -- val mse: 18.22908 -- round 5
00:00:00 -- train mse: 17.79469 -- val mse: 17.33769 -- round 6
00:00:01 -- train mse: 17.62692 -- val mse: 22.67012 -- round 7
00:00:01 -- train mse: 17.24799 -- val mse: 22.77802 -- round 8