Tree API¶

class msitrees.tree.MSIDecisionTreeClassifier(**kwargs)¶

MSI Decision Tree Classifier

Based on breadth-first tree traversal, this no-hyperparameter tree building algorithm tries to create new decision nodes by performing temporary split for each candidate node (at any point of time all current leaves are considered candidate) one by one, and keeping one which decreases overall cost function the most. New branches created with this operation are added to candidate pool. Best split points are estimated with gini based information gain. Training ends when any new split would only add needless complexity to the tree.

Cost function follows paper implementation [1] based on harmonic mean of model inaccuracy and surfeit, but with modifications to approximation of I(X, M) for performance and reusability reasons.

root¶

Root of a decision tree. All decision and leaf nodes are children of this node.

Type: MSINode

fitted¶

Boolean variable indicating if tree was previously fitted.

Type: bool

shape¶

Shape of dataset X with (n_samples, n_features) or None if tree was not yet fitted.

Type: tuple

ncls¶

Number of classification categories or None if tree was not yet fitted.

Type: int

ndim¶

Number of dataset X dimensions. 1 if n_features eq 1, 2 if n_features > 1 or None if tree was not yet fitted.

Type: int

importances¶

Array with feature importances or None if tree was not fitted.

Type: np.ndarray

References

Examples

>>> from msitrees.tree import MSIDecisionTreeClassifier
>>> from sklearn.datasets import load_iris
>>> from sklearn.model_selection import cross_val_score
>>> data = load_iris()
>>> clf = MSIDecisionTreeClassifier()
>>> cross_val_score(clf, data['data'], data['target'], cv=10)
...
array([1.        , 1.        , 1.        , 0.93333333, 0.93333333,
    0.8       , 0.93333333, 0.86666667, 0.8       , 1.        ])

property feature_importances_¶

Returns feature importances

Each feature importance is calculated as normalized sum of gini based information gain at nodes where split was made on that particular feature.

Returns: importances – Normalized array of feature importances.
Return type: np.ndarray

fit(x: Union[numpy.ndarray, pandas.core.frame.DataFrame, pandas.core.series.Series], y: Union[numpy.ndarray, pandas.core.series.Series]) → msitrees.tree.MSIDecisionTreeClassifier ¶

Fits decision tree to training dataset.

Parameters

x (np.ndarray) – Training data of shape (n_samples, n_features) or (n_samples, ). All values have to be numerical, so perform any required encoding before calling this method.
y (np.ndarray) – Ground truth data of shape (n_samples, ). All values have to be numerical, so perform any required encoding before calling this method.

Returns

self – Fitted estimator.

Return type

MSIDecisionTreeClassifier

get_depth() → int¶

Returns decision tree depth

Returns: depth – Maximum depth of fitted decision tree.
Return type: int

get_n_leaves() → int¶

Returns number of tree leaves

Returns: num_leaves – Number of leaf nodes in fitted tree.
Return type: int

get_params(deep: bool = True) → dict¶: scikit-learn API compatibility

predict(x: Union[numpy.ndarray, pandas.core.frame.DataFrame]) → numpy.ndarray¶

Predicts class labels for input data X

Parameters: x (np.ndarray) – Array of samples with shape (n_samples, n_features). Class label is predicted for each sample.
Returns: pred – Array with shape (n_samples, ) Class label prediction for each sample.
Return type: np.ndarray

predict_log_proba(x: Union[numpy.ndarray, pandas.core.frame.DataFrame]) → numpy.ndarray¶

Predicts class log probability for input data X.

Probability is defined as fraction of class label in a leaf.

Parameters: x (np.ndarray) – Array of samples with shape (n_samples, n_features). Class log probabilities are predicted for each sample.
Returns: logprobas – Array with shape (n_samples, n_targets) Array of log probabilities. Each index corresponds to class label and holds predicted log porbability of this class.
Return type: np.ndarray

predict_proba(x: Union[numpy.ndarray, pandas.core.frame.DataFrame]) → numpy.ndarray¶

Predicts class probability for input data X.

Probability is defined as fraction of class label in a leaf.

Parameters: x (np.ndarray) – Array of samples with shape (n_samples, n_features). Class probabilities are predicted for each sample.
Returns: probas – Array with shape (n_samples, n_targets) Array of probabilities. Each index corresponds to class label and holds predicted porbability of this class.
Return type: np.ndarray

score(x: numpy.ndarray, y: numpy.ndarray) → float¶

Predicts from X and computes accuracy score wrt. y

Parameters

x (np.ndarray) – Array of samples with shape (n_samples, n_features). Class label is predicted for each sample.
y (np.ndarray) – Array of ground truth labels.

Returns

accuracy – Accuracy score for predicted class labels.

Return type

float

set_params(**params: dict) → msitrees.tree.MSIDecisionTreeClassifier ¶: scikit-learn API compatibility