Crate oxicuda_anomaly 

oxicuda-anomaly — Anomaly Detection primitives for OxiCUDA.

Pure-Rust implementation of canonical anomaly detection algorithms, suitable for CPU simulation and PTX kernel generation for GPU execution.

Architecture

oxicuda-anomaly
├── svdd/           — DeepSVDD (Ruff et al. 2018)
├── reconstruction/ — Autoencoder & VAE anomaly detection
├── distance/       — LOF, k-NN distance scorer
├── density/        — COPOD, Mahalanobis
├── isolation/      — Isolation Forest random-projection scorer
├── forest/         — Robust Random Cut Forest (CoDisp, streaming)
├── streaming/      — xStream (StreamHash + half-space chains)
├── subspace/       — RS-Hash (randomized subspace hashing)
├── statistical/    — MAD, Z-score, percentile threshold
├── ensemble/       — Ensemble scoring (Average / Maximum / Weighted)
├── metrics/        — AUC-ROC, AUC-PR, F1, detection metrics
├── error           — AnomalyError / AnomalyResult
├── handle          — AnomalyHandle (SmVersion + LcgRng)
└── ptx_kernels     — GPU PTX kernel strings (7 kernels × 6 SM versions)

Modules

density

Density-based anomaly detection (COPOD, Mahalanobis, GMM, KDE, FastMCD).

distance

Distance-based anomaly detection (LOF, kNN score, LOF with k-d tree, CBLOF, ABOD, COF, FastABOD, SOD).

ensemble

Ensemble anomaly scoring.

error

Error types for oxicuda-anomaly.

forest

Random-cut-forest anomaly detection.

graph

Graph-based anomaly detection.

handle

Session handle for oxicuda-anomaly.

isolation

Isolation-based anomaly scoring.

metrics

Anomaly detection evaluation metrics.

prelude

Convenience re-exports of the most-used anomaly detection types.

ptx_kernels

PTX GPU kernel sources for anomaly detection operations.

reconstruction

Reconstruction-based anomaly detection (Autoencoder, VAE, PCA reconstruction, DAGMM, Memory-Augmented Autoencoder, Self-Supervised, AnoGAN / f-AnoGAN, Diffusion).

statistical

Statistical anomaly detection methods (MAD, Z-score, percentile threshold, HBOS, ECOD, conformal p-values, concept-drift detectors, online streaming detectors, and Extreme Value Theory GPD detector).

streaming

Streaming anomaly detection for feature-evolving data streams.

subspace

Subspace anomaly detection via randomized hashing.

svdd

Deep SVDD (Ruff et al. 2018) — One-Class Deep Learning.

time_series

Time-series anomaly detection.