HPCSeries Core Documentation
Version: 0.7.0
HPCSeries Core is a high-performance statistical computing library with SIMD vectorization, OpenMP parallelization, and adaptive auto-tuning. Built with Fortran, C, and C++ for maximum performance, with zero-copy Python bindings via Cython.
Key Features
SIMD-Accelerated Operations: Automatic vectorization using AVX2/AVX-512/SSE2
OpenMP Parallelization: Multi-threaded operations for large datasets
Zero-Copy Integration: Direct NumPy array access without data copying
Adaptive Auto-Tuning: Automatic calibration for optimal performance
Robust Statistics: MAD-based outlier detection and robust z-scores
Rolling Operations: Fast sliding window computations
Anomaly Detection: Statistical and robust anomaly detection methods
Performance
2-10x faster than NumPy for common operations
50-100x faster than Pandas for rolling operations
Sub-microsecond latency for small arrays
Scales to billions of elements
Quick Start
Installation:
pip install hpcs
Basic usage:
import hpcs
import numpy as np
# Create sample data
x = np.random.randn(1000000)
# SIMD-accelerated reductions
hpcs.sum(x) # 2-5x faster than np.sum
hpcs.mean(x) # 2-5x faster than np.mean
hpcs.std(x) # 2-5x faster than np.std
# Rolling operations
hpcs.rolling_mean(x, window=50) # 50-100x faster than pandas
hpcs.rolling_median(x, window=100) # 100x faster than pandas
# Robust statistics and anomaly detection
hpcs.median(x)
hpcs.mad(x)
anomalies = hpcs.detect_anomalies(x, threshold=3.0)
Documentation Contents
User Guide
Examples & Tutorials
API Reference
Development