geopfa.extrapolation.get_predictions

get_predictions(model, X, kvals_df=None, Y_mean=None, Y_std=None)[source]

Generate GP predictions, optionally converting back to original Y-units and optionally reshaping predictions into a 2D grid defined by x/y coordinates.

Parameters:

  • model (GPy.models.GPRegression) – Trained GP regression model.

  • X (numpy.ndarray) – Input features in the same standardized space used during training, shape (N, D).

  • kvals_df (pandas.DataFrame or dict, optional) – Must contain 'x' and 'y'. If provided, predictions are reshaped into a pivoted 2-D grid of shape (len(y_unique), len(x_unique))

  • Y_mean (float or None, optional) – Mean of Y from the training dataset (for de-standardizing predictions).

  • Y_std (float or None, optional) – Standard deviation of Y from the training dataset.

Returns:

numpy.ndarray or tuple – If kvals_df is None, returns (Y_pred_flat, Y_std_flat) where both arrays have shape (N,). If kvals_df is provided, returns (Y_pred_grid, Y_std_grid, x_unique, y_unique) where the grids have shape (ny, nx) and the coordinate arrays contain the sorted unique values from the pivot.

Notes

  • This function reproduces the original behavior precisely:

    • No additional sorting is applied beyond pandas pivot ordering.

    • De-standardization (mean/std) is optional and only applied if both parameters are provided.