wehub-resource-sync
48 lines
1.7 KiB
Python
48 lines
1.7 KiB
Python
"""One-call seed initialization for reproducible notebook runs.
|
|
|
|
Notebooks that produce any random output should call ``set_global_seeds()``
|
|
in their preamble, between imports and the first computation. Monte Carlo
|
|
demos that *want* per-run variability should still call it, with the seed
|
|
declared in their parameters cell so readers can change it explicitly.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import os
|
|
import random
|
|
|
|
|
|
def set_global_seeds(seed: int = 42) -> None:
|
|
"""Seed Python ``random``, NumPy, Torch (CPU+CUDA), and ``PYTHONHASHSEED``.
|
|
|
|
Polars and pandas operations that need a seed accept it per-call
|
|
(e.g. ``df.sample(seed=seed)``) — there is no global polars seed to set.
|
|
Scikit-learn estimators take ``random_state=`` per-instance.
|
|
|
|
Returns ``None`` rather than the seed so that a bare
|
|
``set_global_seeds(SEED)`` in a notebook preamble does not render a
|
|
spurious ``42`` execute_result. Notebooks that want to echo the seed back
|
|
to the reader should ``print(SEED)`` explicitly.
|
|
|
|
``PYTHONHASHSEED`` is set on ``os.environ`` for the benefit of subprocesses
|
|
spawned after the call; it has **no effect** on hash randomization in the
|
|
currently running interpreter (that value is read once at startup). For
|
|
end-to-end hash determinism the kernel must be launched with
|
|
``PYTHONHASHSEED=<seed>`` already set in the environment.
|
|
"""
|
|
os.environ["PYTHONHASHSEED"] = str(seed)
|
|
random.seed(seed)
|
|
|
|
import numpy as np
|
|
|
|
np.random.seed(seed)
|
|
|
|
try:
|
|
import torch
|
|
except ImportError:
|
|
pass
|
|
else:
|
|
torch.manual_seed(seed)
|
|
if torch.cuda.is_available():
|
|
torch.cuda.manual_seed_all(seed)
|