This project designs pipeline to evaluate and benchmark various state-of-the-art algorithms for handling drift. The pipeline is extensible, allowing easy integration of new datasets and algorithms. Refer to the project link here [https://ucsc-ospo.github.io/report/osre24/anl/last/20240610-williamn/].
algorithm/: Contains implementations of various drift management algorithms.algorithm.py: Core algorithm implementation.alwaysretrain.py,aue.py,driftsurf.py,matchmaker.py,noretrain.py: Specific algorithms and strategies.
datastream/: Defines different data stream types.circle.py,covcon.py,datastream.py,ioadmission.py,sine.py: Data stream generators.
drift-detector/: Contains drift detection algorithms.alwaysdrift.py,driftdetector.py: Drift detection mechanisms.
metric-evaluator/: Metrics and evaluation tools.metricevaluator.py: Metrics evaluation.
misc/: Miscellaneous files, including large data files.pipeline/: Main pipeline components.prequential.py: Prequential evaluation setup.
results/: Results from pipeline executions.OSRE-DriftBenchmark.ipynb: Jupyter notebook for ease of use and demonstration.
You can use the provided .ipynb notebook to get started with the experiments! Results from prior executions are also available at the /results folder.