alsdb.processing.biomass.wrap_sklearn_model#

alsdb.processing.biomass.wrap_sklearn_model(estimator, features: list[str] | None = None) → Callable[source]#

Wrap a fitted scikit-learn estimator as a model_fn for compute_biomass.

Handles the reshaping between the per-cell metric dict used internally and the (n_samples, n_features) matrix expected by sklearn, and masks NaN pixels so they are never passed to predict().

Parameters:

  • estimator – Any fitted sklearn-compatible estimator that exposes a predict(X) method (e.g. RandomForestRegressor, GradientBoostingRegressor, Pipeline, …).

  • features – Ordered list of metric names to use as model features. Defaults to all sixteen standard metrics (_METRIC_NAMES): height percentiles, canopy cover, density, FHD, VCI, CRR, and the six height-stratum proportions. The order must match the feature order used during training. Pass an explicit list (e.g. ["h50", "h95", "cc"]) when the model was trained on a subset.

Returns:

A function model_fn(metrics) → np.ndarray compatible with the model_fn parameter of compute_biomass.

Return type:

Callable

Examples

from sklearn.ensemble import RandomForestRegressor
from alsdb.processing.biomass import compute_biomass, wrap_sklearn_model

rf = RandomForestRegressor(n_estimators=200)
rf.fit(X_train, y_train)  # X columns must match _METRIC_NAMES order

model_fn = wrap_sklearn_model(rf)
compute_biomass(provider, store, resolution=10.0, year=2021,
                model_fn=model_fn)
On this page