Transformers

ABERRANT supports streaming transformers that compose with models using |.

Preprocessing

from aberrant.transform.preprocessing import MinMaxScaler, StandardScaler

Projection

from aberrant.transform.projection import IncrementalPCA, RandomProjection

Pipeline example

from aberrant.model.iforest import OnlineIsolationForest
from aberrant.transform.preprocessing import StandardScaler

pipeline = StandardScaler() | OnlineIsolationForest(window_size=512)

for x in stream_of_feature_dicts:
    pipeline.learn_one(x)
    score = pipeline.score_one(x)

Notes

• Keep feature keys consistent across the stream.
• HalfSpaceTrees works best after min-max scaling to [0, 1].