hummingbird.ml.operator_converters.sklearn.iforest

Converters for scikit-learn isolation forest.

class hummingbird.ml.operator_converters.sklearn.iforest.GEMMIsolationForestImpl(*args: Any, **kwargs: Any)

Bases: GEMMTreeImpl

Class implementing the GEMM strategy (in PyTorch) for isolation forest model.

_abc_impl = <_abc_data object>

aggregation(x)

Method defining the aggregation operation to execute after the model is evaluated.

Args:

x: An input tensor

Returns:

The tensor result of the aggregation

class hummingbird.ml.operator_converters.sklearn.iforest.PerfectTreeTraversalIsolationForestImpl(*args: Any, **kwargs: Any)

Bases: PerfectTreeTraversalTreeImpl

Class implementing the Perfect Tree Traversal strategy in PyTorch for isolation forest model.

_abc_impl = <_abc_data object>

aggregation(x)

Method defining the aggregation operation to execute after the model is evaluated.

Args:

x: An input tensor

Returns:

The tensor result of the aggregation

class hummingbird.ml.operator_converters.sklearn.iforest.TreeTraversalIsolationForestImpl(*args: Any, **kwargs: Any)

Bases: TreeTraversalTreeImpl

Class implementing the Tree Traversal strategy in PyTorch for isolation forest model.

_abc_impl = <_abc_data object>

aggregation(x)

Method defining the aggregation operation to execute after the model is evaluated.

Args:

x: An input tensor

Returns:

The tensor result of the aggregation

hummingbird.ml.operator_converters.sklearn.iforest._average_path_length(n_samples_leaf)

Taken from sklearn implementation of isolation forest: https://github.com/scikit-learn/scikit-learn/blob/fd237278e/sklearn/ensemble/_iforest.py#L480 For each given number of samples in the array n_samples_leaf, this calculates average path length of unsucceesful BST search.

Args:

n_samples_leaf: array of number of samples (in leaf)

Returns:

array of average path lengths

hummingbird.ml.operator_converters.sklearn.iforest._get_iforest_anomaly_score_per_node(children_left, children_right, n_node_samples)

Get anomaly score per node in isolation forest, which is node depth + _average_path_length(n_node_samples). Will be used to replace “value” in each tree.

Args:

children_left: left children children_right: right children n_node_samples: number of samples per node

hummingbird.ml.operator_converters.sklearn.iforest._get_parameters_for_sklearn_iforest(tree_infos, extra_config)

Parse sklearn-based isolation forest, replace existing values of node with anomaly score calculated in _get_iforest_anomaly_score_per_node

Args:

tree_infos: The information representing a tree (ensemble)

Returns:

The tree parameters wrapped into an instance of operator_converters._tree_commons_TreeParameters

hummingbird.ml.operator_converters.sklearn.iforest.convert_sklearn_isolation_forest(operator, device, extra_config)

Converter for sklearn.ensemble.IsolationForest.

Args:

operator: An operator wrapping a tree (ensemble) isolation forest model device: String defining the type of device the converted operator should be run on extra_config: Extra configuration used to select the best conversion strategy

Returns:

A PyTorch model