CapyMOA trees wrappers.
Incremental (Hoeffding-bound based) decision trees backed by CapyMOA/MOA. Registered under
capymoa. hub keys, mirroring the native TuiML trees family.Classes
Hoeffding Tree (VFDT) incremental classifier (hub key capymoa.HoeffdingTree).
Wraps
HoeffdingTree. Grows a decision tree one instance at a time, using the Hoeffding bound to decide when enough instances have been seen to commit to a split — no need to store or revisit past data.
Constructor
__init__( self, grace_period: int = 200, )
Parameters
grace_period
int
= 200
Number of instances a leaf must observe between split attempts. Smaller values adapt faster but cost more computation.
Attributes
schema_
capymoa.stream.Schema
Stream schema derived from the training data.
learner_
capymoa.classifier.HoeffdingTree
The fitted backing CapyMOA learner.
Notes
Requires the optional CapyMOA extra:
pip install 'tuiml[capymoa]'. The classic single-model streaming learner; assumes a mostly stationary stream. For drifting streams prefer HoeffdingAdaptiveTree or AdaptiveRandomForest.See Also
python
>>> from tuiml.capymoa import HoeffdingTree
>>> from tuiml.datasets.generators import Agrawal
>>> data = Agrawal(n_samples=1000, function=2, random_state=42).generate()
>>> model = HoeffdingTree(grace_period=100).fit(data.X, data.y)
>>> model.predict(data.X[:5]).shape
(5,)Hoeffding Adaptive Tree (HAT) drift-aware classifier (hub key capymoa.HoeffdingAdaptiveTree).
Wraps
HoeffdingAdaptiveTree. Extends the Hoeffding Tree with ADWIN drift detectors at internal nodes: when a subtree's accuracy degrades, an alternate subtree is grown in the background and swapped in once it performs better.
Constructor
__init__( self, grace_period: int = 200, )
Parameters
grace_period
int
= 200
Number of instances a leaf must observe between split attempts.
Attributes
schema_
capymoa.stream.Schema
Stream schema derived from the training data.
learner_
capymoa.classifier.HoeffdingAdaptiveTree
The fitted backing CapyMOA learner.
Notes
Requires the optional CapyMOA extra:
pip install 'tuiml[capymoa]'. Prefer this over the plain HoeffdingTree when the data distribution is expected to change over time.
python
>>> from tuiml.capymoa import HoeffdingAdaptiveTree
>>> from tuiml.datasets.generators import Hyperplane
>>> data = Hyperplane(n_samples=1000, n_drift_features=2,
... random_state=42).generate()
>>> model = HoeffdingAdaptiveTree().fit(data.X, data.y)
>>> model.predict(data.X[:5]).shape
(5,)Extremely Fast Decision Tree classifier (hub key capymoa.EFDT).
Wraps
EFDT. A Hoeffding Tree variant that splits as soon as a split looks useful rather than provably best, then keeps re-evaluating and revising splits as more data arrives — converging to the batch tree faster than VFDT.
Constructor
__init__( self, grace_period: int = 200, )
Parameters
grace_period
int
= 200
Number of instances a leaf must observe between split attempts.
Attributes
schema_
capymoa.stream.Schema
Stream schema derived from the training data.
learner_
capymoa.classifier.EFDT
The fitted backing CapyMOA learner.
Notes
Requires the optional CapyMOA extra:
pip install 'tuiml[capymoa]'.See Also
python
>>> from tuiml.capymoa import EFDT
>>> from tuiml.datasets.generators import Agrawal
>>> data = Agrawal(n_samples=1000, function=3, random_state=42).generate()
>>> model = EFDT().fit(data.X, data.y)
>>> model.predict(data.X[:5]).shape
(5,)FIMT-DD streaming regression tree (hub key capymoa.FIMTDD).
Wraps
FIMTDD. Fast Incremental Model Tree with Drift Detection: grows a regression tree incrementally, fits linear models in the leaves, and replaces subtrees when drift is detected.
Constructor
__init__( self, )
Attributes
schema_
capymoa.stream.Schema
Stream schema derived from the training data.
learner_
capymoa.regressor.FIMTDD
The fitted backing CapyMOA learner.
Notes
Requires the optional CapyMOA extra:
pip install 'tuiml[capymoa]'.
python
>>> from tuiml.capymoa import FIMTDD
>>> from tuiml.datasets.generators import Friedman
>>> data = Friedman(n_samples=1000, random_state=42).generate()
>>> model = FIMTDD().fit(data.X, data.y)
>>> model.predict(data.X[:5]).shape
(5,)