BUG: BitGenerator.seed_seq doesn't round-trip through pickle #26234
Description
Describe the issue:
The seed_seq of a BitGenerator that's been round-tripped through pickle dump/load does not match with the source. This is despite the fact that the BitGenerator.state (and hence the sequence of variates generated) do round-trip, as does a bare SeedSequence.
This also shows up as a deviation of child bitgens from BitGenerator.spawn()
I'm assuming this is an oversight, since the pickle bytestream generated from a BitGenerator just doesn't mention the seed_seq element at all, and I'm assuming it gets generated from the SeedSequence default initializer by fetching fresh entropy from the OS, which presumably adds a few us delay to the unpickle path.
Reproduce the code example:
import numpy as np
import pickle
bg = np.random.PCG64(np.random.SeedSequence(1))
bg2 = pickle.loads(pickle.dumps(bg))
assert bg.seed_seq == bg2.seed_seqError message:
No response
Python and NumPy Versions:
1.26.4
3.11.8 (main, Feb 26 2024, 15:43:17) [Clang 14.0.6 ]
Runtime Environment:
[{'numpy_version': '1.26.4',
'python': '3.11.8 (main, Feb 26 2024, 15:43:17) [Clang 14.0.6 ]',
'uname': uname_result(system='Darwin', node='LSB-MAC.local', release='22.6.0', version='Darwin Kernel Version 22.6.0: Mon Feb 19 19:48:53 PST 2024; root:xnu-8796.141.3.704.6~1/RELEASE_X86_64', machine='x86_64')},
{'simd_extensions': {'baseline': ['SSE', 'SSE2', 'SSE3'],
'found': ['SSSE3',
'SSE41',
'POPCNT',
'SSE42',
'AVX',
'F16C',
'FMA3',
'AVX2'],
'not_found': ['AVX512F',
'AVX512CD',
'AVX512_KNL',
'AVX512_SKX',
'AVX512_CLX',
'AVX512_CNL',
'AVX512_ICL']}},
{'architecture': 'Haswell',
'filepath': '/usr/local/Caskroom/miniconda/base/envs/rng-seeding/lib/python3.11/site-packages/numpy/.dylibs/libopenblas64_.0.dylib',
'internal_api': 'openblas',
'num_threads': 4,
'prefix': 'libopenblas',
'threading_layer': 'pthreads',
'user_api': 'blas',
'version': '0.3.23.dev'}]
Context for the issue:
The new numpy 1.25 functionality to allow spawning directly from generators (introduced in #23195)rather than needing to pass around a SeedSequence could simplify some multiprocessing/multithreading workflows, but we lose reproducibility due to this bug.