shapiq_benchmark.setups.InterventionalTreeSetup¶

class shapiq_benchmark.setups.InterventionalTreeSetup(*, dataset, random_state=42, dataset_params=<factory>, test_size=0.2, model='decision_tree', preset=None, model_params=<factory>, x=0, n_reference=100, class_index=None, normalize=False)[source]¶

Bases: ModelSetup

An InterventionalTreeGame of a tree model trained on a dataset.

The reference data is a seeded random subset of the training split; the explained point is taken from the test split.

Variables:

model – A tree model name (see PathDependentTreeSetup). Defaults to "decision_tree".
x – The index of the explained point in the test split. Defaults to 0.
n_reference – The number of reference rows. Defaults to 100.
class_index – The explained class for classifiers (None means class 1).
normalize – Whether to center the game. Defaults to False.

Examples

>>> setup = InterventionalTreeSetup(dataset="xor", n_reference=20)
>>> setup.build().reference_data.shape
(20, 4)

build()[source]¶

Train the model and build the game.

Return type:: InterventionalTreeGame

estimator(task, **overrides)¶

Return the unfitted, seeded model for a task.

Parameters:

task (Task) – "classification" or "regression".
**overrides (Any) – Hyperparameters overriding model_params.

Return type:

Any

fit(split)¶

Return the seeded model fitted on the training part of split.

Return type:: Any

load()¶

Load the dataset.

Return type:: Dataset

load_split()¶

Load the dataset and split it (seeded, stratified for classification).

Return type:: DatasetSplit

sample_rows(n_rows, n)¶

Return n sorted row indices out of n_rows, drawn with random_state.

All rows are returned if n is None or at least n_rows.

Return type:: ndarray

stratified_rows(y, n, task)¶

Return n sorted row indices, drawn with random_state and stratified by class.

The rows are stratified for classification where every class has enough rows. All rows are returned if n is None or at least the number of rows.

Return type:: GenericAlias[integer]

to_dict()¶

Return the setup as a JSON-serializable dictionary (the inverse of setup_from_dict()).

Return type:: dict[str, Any]

class_index = None¶

dataset¶

dataset_params¶

property key¶

A stable identifier of the game this setup builds, used as its cache key.

It is a digest of the setup’s name, version, RECIPE_VERSION, and fields (runtime fields excluded), so it is the same across processes and machines.

model = 'decision_tree'¶

model_params¶

n_reference = 100¶

name = 'interventional_tree'¶: The registered name of the setup.

normalize = False¶

preset = None¶

random_state = 42¶

tasks = ('classification', 'regression')¶: The tasks of the datasets the setup accepts.

test_size = 0.2¶

version = 1¶: The version of build(), part of key.

x = 0¶