MIRCO

MIRCO stands for Matrices, Interpolation, Random numbers, Calculus, Optimization. It is the science kit of Hilde, in the place NumPy and SciPy have for Python: a package with the prefix Mirco, under the same licence as Hilde. Source of the language: github.com/heikozim/hilde.

What it holds

• Arrays: NdArray[T] with views and broadcasting, a wire format and a compute layer.
• Linear algebra: LU, Cholesky, QR, least squares, eigenvalues and SVD.
• Interpolation: with splines.
• Optimisation: with Brent, golden section, Nelder-Mead and BFGS.
• Also: FFT, integration and statistics.
• Random numbers: PCG64 and PCG64DXSM; defaultRng gives the same numbers as NumPy for the same seed (SeedSequence, Ziggurat).
• Units: in the type, after SI.
• Files: ROOT of CERN, RNTuple with XXH3 checks, HDF5, netCDF; xz and Zstandard.
• C backend: with BLIS, LAPACK and pocketfft, and calls from Python through ctypes.

"unsafe" only in the bindings

All eight unsafe modules of MIRCO lie in src/Bindings: DuckDB, HDF5, LAPACK, libm, netCDF, Number, pocketfft and Ziggurat. Every other module is safe and is built on them.

Parts of MIRCO

• YOMPI: plots as SVG and PNG.
• SCHLUCKI: tables with named columns.
• SPERLI: parallel jobs over the network, in Hilde's standard library, aligned with MIRCO's array type.