splittable_random v0.17.0 · OCaml Package

A splittable pseudo-random number generator (SPRNG) functions like a PRNG in that it can be used as a stream of random values; it can also be "split" to produce a second, independent stream of random values.

This module implements a splittable pseudo-random number generator that sacrifices cryptographic-quality randomness in favor of performance.

The primary difference between Splittable_random and Random is the split operation for generating new pseudo-random states. While it is easy to simulate split using Random, the result has undesirable statistical properties; the new state does not behave independently of the original. It is better to switch to Splittable_random if you need an operation like split, as this module has been implemented with the statistical properties of splitting in mind. For most other purposes, Random is likely a better choice, as its implementation passes all Diehard tests, while Splittable_random fails some Diehard tests.

type t

val create : Base.Random.State.t -> t

Create a new t seeded from the given random state. This allows nondeterministic initialization, for example in the case that the input state was created using Random.make_self_init.

Constructors like create and of_int should be called once at the start of a randomized computation and the resulting state should be threaded through. Repeatedly creating splittable random states from seeds in the middle of computation can defeat the SPRNG's splittable properties.

val of_int : Base.int -> t

Create a new t that will return identical results to any other t created with that integer.

val perturb : t -> Base.int -> Base.unit

perturb t salt adds the entropy of salt to t.

val copy : t -> t

Create a copy of t that will return the same random samples as t.

val split : t -> t

split t produces a new state that behaves deterministically (i.e. only depending on the state of t), but pseudo-independently from t. This operation mutates t, i.e., t will return different values than if this hadn't been called.

module State : sig ... end

Legacy aliases for the preceding definitions.

val bool : t -> Base.bool

Produces a random, fair boolean.

val int : t -> lo:Base.int -> hi:Base.int -> Base.int

Produce a random number uniformly distributed in the given inclusive range. (In the case of float, hi may or may not be attainable, depending on rounding.)

val int32 : t -> lo:Base.int32 -> hi:Base.int32 -> Base.int32

val int63 : t -> lo:Base.Int63.t -> hi:Base.Int63.t -> Base.Int63.t

val int64 : t -> lo:Base.int64 -> hi:Base.int64 -> Base.int64

val nativeint : t -> lo:Base.nativeint -> hi:Base.nativeint -> Base.nativeint

val float : t -> lo:Base.float -> hi:Base.float -> Base.float

val unit_float : t -> Base.float

unit_float state = float state ~lo:0. ~hi:1., but slightly more efficient (and right endpoint is exclusive).

module Log_uniform : sig ... end

package splittable_random