AdaptiveRandomForestRegressor

Adaptive Random Forest regressor.

The 3 most important aspects of Adaptive Random Forest ¹ are:

1. inducing diversity through re-sampling

2. inducing diversity through randomly selecting subsets of features for node splits

3. drift detectors per base tree, which cause selective resets in response to drifts

Notice that this implementation is slightly different from the original algorithm proposed in ². The HoeffdingTreeRegressor is used as base learner, instead of FIMT-DD. It also adds a new strategy to monitor the predictions and check for concept drifts. The deviations of the predictions to the target are monitored and normalized in the [0, 1] range to fulfill ADWIN's requirements. We assume that the data subjected to the normalization follows a normal distribution, and thus, lies within the interval of the mean \(\pm3\sigma\).

Parameters

• n_models (int) – defaults to 10

  Number of trees in the ensemble.

• max_features – defaults to sqrt

  Max number of attributes for each node split.
  - If int, then consider max_features at each split.
  - If float, then max_features is a percentage and int(max_features * n_features) features are considered per split.
  - If "sqrt", then max_features=sqrt(n_features).
  - If "log2", then max_features=log2(n_features).
  - If None, then max_features=n_features.

• aggregation_method (str) – defaults to median

  The method to use to aggregate predictions in the ensemble.
  - 'mean'
  - 'median' - If selected will disable the weighted vote.

• lambda_value (int) – defaults to 6

  The lambda value for bagging (lambda=6 corresponds to Leveraging Bagging).

• metric (river.metrics.base.RegressionMetric) – defaults to MSE: 0.

  Metric used to track trees performance within the ensemble. Depending, on the configuration, this metric is also used to weight predictions from the members of the ensemble.

• disable_weighted_vote – defaults to True

  If True, disables the weighted vote prediction, i.e. does not assign weights to individual tree's predictions and uses the arithmetic mean instead. Otherwise will use the metric value to weight predictions.

• drift_detector (base.DriftDetector) – defaults to ADWIN

  Drift Detection method. Set to None to disable Drift detection.

• warning_detector (base.DriftDetector) – defaults to ADWIN

  Warning Detection method. Set to None to disable warning detection.

• grace_period (int) – defaults to 50

  [Tree parameter] Number of instances a leaf should observe between split attempts.

• max_depth (int) – defaults to None

  [Tree parameter] The maximum depth a tree can reach. If None, the tree will grow indefinitely.

• delta (float) – defaults to 0.01

  [Tree parameter] Allowed error in split decision, a value closer to 0 takes longer to decide.

• tau (float) – defaults to 0.05

  [Tree parameter] Threshold below which a split will be forced to break ties.

• leaf_prediction (str) – defaults to adaptive

  [Tree parameter] Prediction mechanism used at leaves.
  - 'mean' - Target mean
  - 'model' - Uses the model defined in leaf_model
  - 'adaptive' - Chooses between 'mean' and 'model' dynamically
  

• leaf_model (base.Regressor) – defaults to None

  [Tree parameter] The regression model used to provide responses if leaf_prediction='model'. If not provided, an instance of river.linear_model.LinearRegression with the default hyperparameters is used.

• model_selector_decay (float) – defaults to 0.95

  [Tree parameter] The exponential decaying factor applied to the learning models' squared errors, that are monitored if leaf_prediction='adaptive'. Must be between 0 and 1. The closer to 1, the more importance is going to be given to past observations. On the other hand, if its value approaches 0, the recent observed errors are going to have more influence on the final decision.

• nominal_attributes (list) – defaults to None

  [Tree parameter] List of Nominal attributes. If empty, then assume that all attributes are numerical.

• splitter (river.tree.splitter.base.Splitter) – defaults to None

  [Tree parameter] The Splitter or Attribute Observer (AO) used to monitor the class statistics of numeric features and perform splits. Splitters are available in the tree.splitter module. Different splitters are available for classification and regression tasks. Classification and regression splitters can be distinguished by their property is_target_class. This is an advanced option. Special care must be taken when choosing different splitters.By default, tree.splitter.EBSTSplitter is used if splitter is None.

• min_samples_split (int) – defaults to 5

  [Tree parameter] The minimum number of samples every branch resulting from a split candidate must have to be considered valid.

• binary_split (bool) – defaults to False

  [Tree parameter] If True, only allow binary splits.

• max_size (float) – defaults to 500.0

  [Tree parameter] Maximum memory (MB) consumed by the tree.

• memory_estimate_period (int) – defaults to 2000000

  [Tree parameter] Number of instances between memory consumption checks.

• stop_mem_management (bool) – defaults to False

  [Tree parameter] If True, stop growing as soon as memory limit is hit.

• remove_poor_attrs (bool) – defaults to False

  [Tree parameter] If True, disable poor attributes to reduce memory usage.

• merit_preprune (bool) – defaults to True

  [Tree parameter] If True, enable merit-based tree pre-pruning.

• seed (int) – defaults to None

  Random seed for reproducibility.

Attributes

• models

• valid_aggregation_method

  Valid aggregation_method values.

Examples

>>> from river import datasets
>>> from river import evaluate
>>> from river import metrics
>>> from river import ensemble
>>> from river import preprocessing

>>> dataset = datasets.TrumpApproval()

>>> model = (
...     preprocessing.StandardScaler() |
...     ensemble.AdaptiveRandomForestRegressor(seed=42)
... )

>>> metric = metrics.MAE()

>>> evaluate.progressive_val_score(dataset, model, metric)
MAE: 0.800649

Methods

append

S.append(value) -- append value to the end of the sequence

Parameters

• item

clear

S.clear() -> None -- remove all items from S

copy

count

S.count(value) -> integer -- return number of occurrences of value

Parameters

• item

extend

S.extend(iterable) -- extend sequence by appending elements from the iterable

Parameters

• other

index

S.index(value, [start, [stop]]) -> integer -- return first index of value. Raises ValueError if the value is not present.

Supporting start and stop arguments is optional, but recommended.

Parameters

• item
• args

insert

S.insert(index, value) -- insert value before index

Parameters

• i
• item

learn_one

pop

S.pop([index]) -> item -- remove and return item at index (default last). Raise IndexError if list is empty or index is out of range.

Parameters

• i – defaults to -1

predict_one

Predict the output of features x.

Parameters

• x (dict)

Returns

Number: The prediction.

remove

S.remove(value) -- remove first occurrence of value. Raise ValueError if the value is not present.

Parameters

• item

reverse

S.reverse() -- reverse IN PLACE

sort

References

---

1. Gomes, H.M., Bifet, A., Read, J., Barddal, J.P., Enembreck, F., Pfharinger, B., Holmes, G. and Abdessalem, T., 2017. Adaptive random forests for evolving data stream classification. Machine Learning, 106(9-10), pp.1469-1495. ↩

2. Gomes, H.M., Barddal, J.P., Boiko, L.E., Bifet, A., 2018. Adaptive random forests for data stream regression. ESANN 2018. ↩