diff options
| author | Robert Kern <robert.kern@gmail.com> | 2021-05-04 17:09:39 -0400 |
|---|---|---|
| committer | Robert Kern <robert.kern@gmail.com> | 2021-05-04 17:09:39 -0400 |
| commit | 32ebda5a11e30e1a57a3ad3bf1798ca651fb570f (patch) | |
| tree | ddb471f9e145201a38ab0b7d7c8b9d6ef2cdf4f6 /doc/source/reference/random | |
| parent | 819433652f344526b36cf42b2fadf69e4af8eb45 (diff) | |
| download | numpy-32ebda5a11e30e1a57a3ad3bf1798ca651fb570f.tar.gz | |
DOC: nuance recommendations for PCG64.
Diffstat (limited to 'doc/source/reference/random')
| -rw-r--r-- | doc/source/reference/random/parallel.rst | 9 | ||||
| -rw-r--r-- | doc/source/reference/random/performance.rst | 9 |
2 files changed, 12 insertions, 6 deletions
diff --git a/doc/source/reference/random/parallel.rst b/doc/source/reference/random/parallel.rst index 721584014..7e8758133 100644 --- a/doc/source/reference/random/parallel.rst +++ b/doc/source/reference/random/parallel.rst @@ -88,10 +88,11 @@ territory ([2]_). estimate the naive upper bound on a napkin and take comfort knowing that the probability is actually lower. -.. [2] In this calculation, we can ignore the amount of numbers drawn from each - stream. Each of the PRNGs we provide has some extra protection built in +.. [2] In this calculation, we can mostly ignore the amount of numbers drawn from each + stream. See :ref:`upgrading-pcg64` for the technical details about + `PCG64`. The other PRNGs we provide have some extra protection built in that avoids overlaps if the `~SeedSequence` pools differ in the - slightest bit. `PCG64` has :math:`2^{127}` separate cycles + slightest bit. `PCG64DXSM` has :math:`2^{127}` separate cycles determined by the seed in addition to the position in the :math:`2^{128}` long period for each cycle, so one has to both get on or near the same cycle *and* seed a nearby position in the cycle. @@ -156,6 +157,8 @@ are listed below. +-----------------+-------------------------+-------------------------+-------------------------+ | PCG64 | :math:`2^{128}` | :math:`~2^{127}` ([3]_) | 64 | +-----------------+-------------------------+-------------------------+-------------------------+ +| PCG64DXSM | :math:`2^{128}` | :math:`~2^{127}` ([3]_) | 64 | ++-----------------+-------------------------+-------------------------+-------------------------+ | Philox | :math:`2^{256}` | :math:`2^{128}` | 64 | +-----------------+-------------------------+-------------------------+-------------------------+ diff --git a/doc/source/reference/random/performance.rst b/doc/source/reference/random/performance.rst index 74dad4cc3..812c719f8 100644 --- a/doc/source/reference/random/performance.rst +++ b/doc/source/reference/random/performance.rst @@ -5,9 +5,12 @@ Performance Recommendation ************** -The recommended generator for general use is `PCG64`. It is -statistically high quality, full-featured, and fast on most platforms, but -somewhat slow when compiled for 32-bit processes. + +The recommended generator for general use is `PCG64` or its upgraded variant +`PCG64DXSM` for heavily-parallel use cases. They are statistically high quality, +full-featured, and fast on most platforms, but somewhat slow when compiled for +32-bit processes. See :ref:`upgrading-pcg64` for details on when heavy +parallelism would indicate using `PCG64DXSM`. `Philox` is fairly slow, but its statistical properties have very high quality, and it is easy to get assuredly-independent stream by using |