NumpySeedSequence Class

public sealed class NumpySeedSequence : SeedSequence

Public NotInheritable Class NumpySeedSequence
	Inherits SeedSequence

public ref class NumpySeedSequence sealed : public SeedSequence

[<SealedAttribute>]
type NumpySeedSequence = 
    class
        inherit SeedSequence
    end

Inheritance

Object  →  SeedSequence  →  NumpySeedSequence

NumpySeedSequence provides byte-for-byte compatibility with NumPy's numpy.random.SeedSequence class, enabling deterministic seeding that produces identical results across Python (NumPy) and .NET when using the same seed values and stream identifiers.

Algorithm:

The algorithm uses a pool-based mixing design derived from Melissa O'Neill's improved seed_seq design. Input entropy is absorbed into a fixed-size pool (default 4 words) using hash mixing operations with constants INIT_A/MULT_A and INIT_B/MULT_B. The pool is then mixed thoroughly to ensure all bits affect all other bits. Entropy generation cycles through the pool, applying per-position mixing to produce output values.

Pool-Based State:

Unlike SplitMix64-based sequences, NumPy's algorithm maintains a pool of 32-bit words (default size: 4 words) that represents the fully mixed entropy. Output generation cycles through this pool, applying position-dependent mixing to each value. The pool size is fixed at construction and determines the period before cycling.

Usage:

This sequence is selected by Numpy. For explicitly supported generators (PCG64, PCG64-DXSM, SFC64, Philox), it provides full NumPy state compatibility. For other generators, it provides NumPy-compatible entropy expansion.

Reference:

NumPy documentation: https://numpy.org/doc/stable/reference/random/bit_generators/ generated/numpy.random.SeedSequence.html

Melissa O'Neill's PCG paper (Appendix on seeding): https://www.pcg-random.org/paper.html

Reference