neural_trees.NumpySoftTree

class neural_trees.NumpySoftTree(depth: int, weights, biases, log_beta, node_logits, is_split, classes: Sequence[Any], feature_names: Sequence[str] | None = None, kind: str = 'classifier', y_mean=None, y_scale=None, single_output: bool = True)

Bases: object

Torch-free copy of a fitted SoftDecisionTree or SoftDecisionTreeRegressor (kind).

Parameters are the tree’s own: gate weights (n_internal, n_features), gate biases, per-gate log temperatures, node logits (n_nodes, n_classes) and the is_split flags that say which internal nodes act as leaves.

predict_values(X) → ndarray

Regressor only: arrival-weighted leaf values, in target units.

save_onnx(path: str, opset: int = 17) → None

Write to_onnx() to a file.

to_onnx(opset: int = 17)

The same model as an onnx.ModelProto; see neural_trees.decision_trees.onnx_export.