Rolling and windowed operations
Rolling sum/mean/variance and related windowed statistics for high-throughput time series.
Open-source • CPU-first • Stable C ABI
Fast, dependable numeric kernels for large-scale computation.
HPCSeries is an open-source engine series built around a focused numeric core: modern Fortran kernels, a stable C interface, and a performance-oriented approach to fundamental primitives.
Designed for scientific computing, data engineering, and performance-critical analytics.
About HPCSeries
A family of engines centered on a reusable numeric core. The core stays domain-agnostic: arrays, indices, groups, and numeric transforms—no application logic embedded.
Stable C ABI
ISO_C_BINDING interfaces with explicit status codes and caller-allocated output buffers.
Performance discipline
CPU-first, cache-aware primitives with benchmarks-driven iteration.
Composable building blocks
Use kernels as the foundation for domain engines without locking into a monolithic framework.
Core capabilities
Focused primitives with predictable performance characteristics and clean interoperability.
Reductions and grouped reductions
Fast reductions plus group-by style aggregations over index groups.
Transforms and robust statistics
Normalization, z-scores, anomaly transforms, and robust variants for noisy signals.
Downloads
Install via pip or download from GitHub.
Get started
Quick start guide with examples and comprehensive API documentation.