Simon Brugman
|
34664a0ca0
|
[`numpy`] numpy-legacy-random (#2960)
The new `Generator` in NumPy uses bits provided by [PCG64](https://numpy.org/doc/stable/reference/random/bit_generators/pcg64.html#numpy.random.PCG64) which has better statistical properties than the legacy [MT19937](https://numpy.org/doc/stable/reference/random/bit_generators/mt19937.html#numpy.random.MT19937) used in [RandomState](https://numpy.org/doc/stable/reference/random/legacy.html#numpy.random.RandomState). Global random functions can also be problematic with parallel processing.
This rule is probably quite useful for data scientists (perhaps in combination with `nbqa`)
References:
- [Legacy Random Generation](https://numpy.org/doc/stable/reference/random/legacy.html#legacy)
- [Random Sampling](https://numpy.org/doc/stable/reference/random/index.html#random-quick-start)
- [Using PyTorch + NumPy? You're making a mistake.](https://tanelp.github.io/posts/a-bug-that-plagues-thousands-of-open-source-ml-projects/)
|
2023-02-17 02:06:30 +00:00 |
Charlie Marsh