API Reference / weka /

rules.py

Weka rule-learner wrappers.

Rule and decision-table learners from weka.classifiers.rules, registered under weka. hub keys. These are the models to reach for when the result has to be readable by a person: every one of them can be printed with to_weka_string().

Notes

Requires the optional Weka extra: pip install 'tuiml[weka]' plus a Java runtime (11+) on PATH.

Classes

ZeroR

class weka.rules.ZeroR(_WekaSupervisedMixin, Classifier)

ZeroR — majority-class baseline (hub key weka.ZeroR).

Wraps weka.classifiers.rules.ZeroR. Predicts the majority class (or the mean, for a numeric target) and ignores every attribute. Its only real use is as the floor any serious model must beat.
Constructor 
__init__(
    self,
    nominal_features: Optional[Sequence[int]] = None,
    options: Optional[Sequence[str]] = None,
)

Parameters

nominal_features
sequence of int or None = None
0-based indices of columns to treat as categorical.
options
sequence of str or None = None
Extra raw Weka options.

See Also

ZeroRuleClassifier

TuiML's native ZeroR.
python 
>>> from tuiml.weka import ZeroR
>>> from tuiml.datasets import load_iris
>>> data = load_iris()
>>> clf = ZeroR().fit(data.X, data.y)
>>> clf.predict(data.X[:5]).shape
(5,)

Methods

get_parameter_schema (cls) -> Dict[str, Any]

Return JSON Schema for constructor parameters.

get_capabilities (cls) -> List[str]

Return the capability names this wrapper supports.

OneR

class weka.rules.OneR(_WekaSupervisedMixin, Classifier)

OneR — one-attribute rule classifier (hub key weka.OneR).

Wraps weka.classifiers.rules.OneR. Builds a single rule on the one attribute that yields the lowest training error — Holte's demonstration that very simple rules often perform surprisingly well.
Constructor 
__init__(
    self,
    min_bucket_size: int = 6,
    nominal_features: Optional[Sequence[int]] = None,
    options: Optional[Sequence[str]] = None,
)

Parameters

min_bucket_size
int = 6
Minimum bucket size used when discretizing numeric attributes (Weka -B). Larger values give simpler, more general rules.
nominal_features
sequence of int or None = None
0-based indices of columns to treat as categorical.
options
sequence of str or None = None
Extra raw Weka options.

See Also

DecisionStumpClassifier

TuiML's native single-attribute rule learner.
python 
>>> from tuiml.weka import OneR
>>> from tuiml.datasets import load_iris
>>> data = load_iris()
>>> clf = OneR(min_bucket_size=6).fit(data.X, data.y)
>>> clf.predict(data.X[:5]).shape
(5,)

Methods

get_parameter_schema (cls) -> Dict[str, Any]

Return JSON Schema for constructor parameters.

get_capabilities (cls) -> List[str]

Return the capability names this wrapper supports.

JRip

class weka.rules.JRip(_WekaSupervisedMixin, Classifier)

JRip — Weka's RIPPER rule learner (hub key weka.JRip).

Wraps weka.classifiers.rules.JRip, Weka's implementation of Cohen's RIPPER: grow a rule set, prune it, then optimize it over several passes.
Constructor 
__init__(
    self,
    folds: int = 3,
    min_no: float = 2.0,
    optimizations: int = 2,
    seed: int = 1,
    prune: bool = True,
    nominal_features: Optional[Sequence[int]] = None,
    options: Optional[Sequence[str]] = None,
)

Parameters

folds
int = 3
Folds used for reduced-error pruning; one is held out (Weka -F).
min_no
float = 2.0
Minimum total instance weight covered by a rule (Weka -N).
optimizations
int = 2
Number of optimization passes over the rule set (Weka -O).
seed
int = 1
Random seed (Weka -S).
prune
bool = True
Run pruning. Passing False emits Weka's -P flag to disable it.
nominal_features
sequence of int or None = None
0-based indices of columns to treat as categorical.
options
sequence of str or None = None
Extra raw Weka options.

See Also

DecisionTreeClassifier

TuiML's nearest native model — axis-aligned rules as a tree.
python 
>>> from tuiml.weka import JRip
>>> from tuiml.datasets import load_iris
>>> data = load_iris()
>>> clf = JRip().fit(data.X, data.y)
>>> clf.predict(data.X[:5]).shape
(5,)

Methods

get_parameter_schema (cls) -> Dict[str, Any]

Return JSON Schema for constructor parameters.

get_capabilities (cls) -> List[str]

Return the capability names this wrapper supports.

PART

class weka.rules.PART(_WekaSupervisedMixin, Classifier)

PART — partial decision tree rule learner (hub key weka.PART).

Wraps weka.classifiers.rules.PART. Repeatedly builds a partial C4.5 tree and turns its best leaf into a rule, which avoids RIPPER's global optimization step while still producing a compact rule list.
Constructor 
__init__(
    self,
    confidence_factor: float = 0.25,
    min_num_obj: int = 2,
    unpruned: bool = False,
    binary_splits: bool = False,
    nominal_features: Optional[Sequence[int]] = None,
    options: Optional[Sequence[str]] = None,
)

Parameters

confidence_factor
float = 0.25
Confidence threshold for pruning (Weka -C).
min_num_obj
int = 2
Minimum instances per rule (Weka -M).
unpruned
bool = False
Build unpruned rules (Weka -U).
binary_splits
bool = False
Use binary splits on nominal attributes (Weka -B).
nominal_features
sequence of int or None = None
0-based indices of columns to treat as categorical.
options
sequence of str or None = None
Extra raw Weka options.

See Also

DecisionTreeClassifier

TuiML's nearest native model — PART's rules come from partial trees.
python 
>>> from tuiml.weka import PART
>>> from tuiml.datasets import load_iris
>>> data = load_iris()
>>> clf = PART().fit(data.X, data.y)
>>> clf.predict(data.X[:5]).shape
(5,)

Methods

get_parameter_schema (cls) -> Dict[str, Any]

Return JSON Schema for constructor parameters.

get_capabilities (cls) -> List[str]

Return the capability names this wrapper supports.

DecisionTable

class weka.rules.DecisionTable(_WekaSupervisedMixin, Classifier)

DecisionTable — simple decision table majority classifier (hub key weka.DecisionTable).

Wraps weka.classifiers.rules.DecisionTable. Selects a feature subset by search, then predicts using a lookup table of majority classes over that subset, falling back to the global majority for unseen combinations.
Constructor 
__init__(
    self,
    cross_val: int = 1,
    use_ibk: bool = False,
    nominal_features: Optional[Sequence[int]] = None,
    options: Optional[Sequence[str]] = None,
)

Parameters

cross_val
int = 1
Cross-validation setting used to evaluate subsets (Weka -X); 1 means leave-one-out.
use_ibk
bool = False
Fall back to a nearest-neighbour prediction rather than the global majority for combinations missing from the table (Weka -I).
nominal_features
sequence of int or None = None
0-based indices of columns to treat as categorical.
options
sequence of str or None = None
Extra raw Weka options.
python 
>>> from tuiml.weka import DecisionTable
>>> from tuiml.datasets import load_iris
>>> data = load_iris()
>>> clf = DecisionTable().fit(data.X, data.y)
>>> clf.predict(data.X[:5]).shape
(5,)

Methods

get_parameter_schema (cls) -> Dict[str, Any]

Return JSON Schema for constructor parameters.

get_capabilities (cls) -> List[str]

Return the capability names this wrapper supports.

M5Rules

class weka.rules.M5Rules(_WekaSupervisedMixin, Regressor)

M5Rules — regression rules from M5 model trees (hub key weka.M5Rules).

Wraps weka.classifiers.rules.M5Rules. Repeatedly builds an M5 model tree and converts its best branch into a rule, giving a rule list whose consequents are linear models.
Constructor 
__init__(
    self,
    min_num_instances: float = 4.0,
    unpruned: bool = False,
    use_unsmoothed: bool = False,
    nominal_features: Optional[Sequence[int]] = None,
    options: Optional[Sequence[str]] = None,
)

Parameters

min_num_instances
float = 4.0
Minimum instances allowed at a leaf (Weka -M).
unpruned
bool = False
Build unpruned rules (Weka -N).
use_unsmoothed
bool = False
Skip the smoothing step (Weka -U).
nominal_features
sequence of int or None = None
0-based indices of columns to treat as categorical.
options
sequence of str or None = None
Extra raw Weka options.
python 
>>> from tuiml.weka import M5Rules
>>> import numpy as np
>>> rng = np.random.default_rng(0)
>>> X = rng.normal(size=(80, 3))
>>> y = 2.0 * X[:, 0] + rng.normal(scale=0.1, size=80)
>>> reg = M5Rules().fit(X, y)
>>> reg.predict(X[:5]).shape
(5,)

Methods

get_parameter_schema (cls) -> Dict[str, Any]

Return JSON Schema for constructor parameters.

get_capabilities (cls) -> List[str]

Return the capability names this wrapper supports.