Proposal to accept NEP 19: Random Number Generator Policy

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Proposal to accept NEP 19: Random Number Generator Policy

Robert Kern-2
I propose that we accept NEP 19: Random Number Generator Policy:


The discussions on this NEP were productive and led to a major revision in how stable random number streams will be maintained in the future for the purpose of unit testing. The current version is much more elegant and imposes less pain on downstream projects. I'd like to thank everyone who participated in that discussion: you really did a good job at clarifying the issues. The NEP is much better because of your input.

If there are no substantive objections within 7 days from this email, then the NEP will be accepted; see NEP 0 for more details:

--
Robert Kern

_______________________________________________
NumPy-Discussion mailing list
[hidden email]
https://mail.python.org/mailman/listinfo/numpy-discussion
Reply | Threaded
Open this post in threaded view
|

Re: Proposal to accept NEP 19: Random Number Generator Policy

Robert Kern-2
There has been one clarification to the text:


For the legacy RandomState, we will *not* be fixing correctness bugs if they doing so would break the stream; this is in contrast with the current policy where we can fix correctness bugs. In the post-NEP world, RandomState's purpose will be to provide across-version-stable numbers for unit tests, so stability is primary. I don't expect to see many more bugs, except in arcane corners of the parameter spaces, like in #11475, which can be avoided, and we can introduce warnings to help users avoid them.


_______________________________________________
NumPy-Discussion mailing list
[hidden email]
https://mail.python.org/mailman/listinfo/numpy-discussion
Reply | Threaded
Open this post in threaded view
|

Re: Proposal to accept NEP 19: Random Number Generator Policy

ralfgommers


On Tue, Jul 3, 2018 at 4:34 PM, Robert Kern <[hidden email]> wrote:
There has been one clarification to the text:


For the legacy RandomState, we will *not* be fixing correctness bugs if they doing so would break the stream; this is in contrast with the current policy where we can fix correctness bugs. In the post-NEP world, RandomState's purpose will be to provide across-version-stable numbers for unit tests, so stability is primary. I don't expect to see many more bugs, except in arcane corners of the parameter spaces, like in #11475, which can be avoided, and we can introduce warnings to help users avoid them.


NEP status is now accept (https://github.com/numpy/numpy/pull/11560). Thanks Robert (& Kevin & @all)!

Ralf



_______________________________________________
NumPy-Discussion mailing list
[hidden email]
https://mail.python.org/mailman/listinfo/numpy-discussion