Add anomaly.RobustRandomCutForest

docs/releases/unreleased.md

@@ -18,6 +18,7 @@
 - Rewrote `anomaly.LocalOutlierFactor`. It now stores samples in a bounded sliding window via a `river.neighbors` search engine (`LazySearch` by default, `SWINN` for approximate search) and computes the LOF of a sample against the window on demand. `learn_one` is now constant-time and memory is bounded by the window size, and `score_one` no longer mutates the model. Scores match scikit-learn over the same window: an unseen point reproduces `LocalOutlierFactor(novelty=True)`, and a stored point reproduces the in-sample `negative_outlier_factor_` (a point is never its own neighbor). `learn_many` now accepts any [narwhals](https://github.com/narwhals-dev/narwhals)-supported eager dataframe (pandas, polars, pyarrow, ...). Behavior changes: scores now reflect the most recent `window_size` samples rather than the entire history, scoring an already-seen point returns its LOF instead of `0.0`, and the `distance_func` parameter is replaced by the engine's distance function.
 
 - Made `anomaly.OneClassSVM.learn_many` dataframe-agnostic via narwhals: it now accepts any narwhals-supported eager backend (pandas, polars, pyarrow, ...) instead of only pandas. Outputs are unchanged.
+- Added `anomaly.RobustRandomCutForest`, an online implementation of the Robust Random Cut Forest (Guha et al., 2016). It maintains an ensemble of robust random cut trees over a sliding window of the stream and scores points by their average collusive displacement (CoDisp); higher scores indicate anomalies. The implementation is dictionary based and `score_one` is side-effect-free.
 
 ## cluster
 

river/anomaly/__init__.py

@@ -20,6 +20,7 @@
 from .loda import LODA
 from .lof import LocalOutlierFactor
 from .pad import PredictiveAnomalyDetection
+from .rrcf import RobustRandomCutForest
 from .sad import StandardAbsoluteDeviation
 from .svm import OneClassSVM
 
@@ -29,6 +30,7 @@
     "LODA",
     "OneClassSVM",
     "QuantileFilter",
+    "RobustRandomCutForest",
     "StandardAbsoluteDeviation",
     "ThresholdFilter",
     "LocalOutlierFactor",