Anomaly Detection and Ranking¶

MPSBoost 1.0.0 includes CPU-suitable anomaly detection and ranking entries.

Isolation Forest¶

Entries:

IsolationForest
MPSIsolationForest

Semantics:

random isolation trees
path length
anomaly score
score_samples
decision_function
predict

This workload is branch-heavy, so CPU is currently expected to be more suitable than Apple GPU. If the user passes device="mps", MPSBoost warns, continues with the CPU backend, and records the reason in training_summary_.

Learning to Rank¶

Entry:

LearningToRankRegressor

Semantics:

group/query input contract
pointwise ranking scorer
full-list NDCG score
sklearn-style parameter protocol

This workflow is latency-sensitive, so CPU is currently expected to be more suitable than Apple GPU. A device="mps" request remains runnable and records the CPU backend decision.