The rationale for this crate is to generate good quality random numbers fast, simply and with a minimal footprint.
Not everyone wants to add 375 kB, plus another ten dependencies, just to generate a bunch of random numbers for testing ( looking at the 'other' crate: rand ).
In contradistinction, this crate is lightweight and it has no dependencies.
Even so, there are four different algorithms on offer, plus a good range of utility functions to easily generate vectors and vectors of vectors filled with random numbers.
The main objective has been the ease of use rather than flat-out speed but the algorithms are neverheless very fast.
It is highly recommended to run tests/tests.rs with examples of usage. Here is a part of it:
rust
println!("20 random bytes: {}",ru8.ranvec(20));
println!("Dice roll: {}",ru8.ranvec_in(20,1.,6.));
println!("5x5 matrix of integers:\n{}",ri.ranvv_in(5,5,-10.,10.));
rust
use ran::*;
These algorithms use thread safe static seeds. It is strongly recommended to initialise them with set_seeds(value); in every thread where you may want to be generating random numbers, otherwise you will get the same sequence every time, based on the default value. Any u64 value will do to initiate a new, different, random sequence. Of course, the same seed will always produce the same sequence but this is actually useful for exact testing comparisons.
```rust /// This function initialises SEED and xoshi seeds X0-X3. /// The supplied value must be non zero, /// otherwise seeds will remain unchanged. pub fn set_seeds( seed:u64 )
/// Resets xoshi seeds without changing the main SEED. /// There is usually no need to reset seeds when they are /// already running. pub fn reset_xoshi() ```
Polymorphic interface avoids having to use different typed functions for each primitive type. This can be repetitive, given that there are quite a few numeric types. Nevertheless, such typed functions are also available in Ran and can be used in simple applications directly (see Explicitly Typed Functions below).
In lib.rs we define three polymorphic (generic types):
```rust /// Wrapper for enum polymorphism - single value pub enum Rnum { F64(f64), U64(u64), I64(i64), U16(u16), U8(u8) // Should be extended to cover all numeric types? }
/// Wrapper for enum polymorphism - vectors
pub enum Rv {
F64(Vec
/// Wrapper for enum polymorphism - vectors of vectors
pub enum Rvv {
F64(Vec
To use the generics, we generate an instance of one of these types. For single random numbers, it will be enum type Rnum, of the variant corresponding to the end-type of the random numbers to be generated. Thus Rnum is just a wrapper, serving to communicate to the generic method(s) information about the actual type of the (random) numbers wanted. The following example shows variables of all the main types being declared:
rust
let rf = Rnum::newf64();
let ru = Rnum::newu64();
let ri = Rnum::newi64();
let ru16 = Rnum::newu16();
let ru8 = Rnum::newu8();
We can then apply common generic method(s) to all such variables to generate the required random numbers. The results are again wrapped inside an Rnum type. For example:
rust
println!("Random numbers in specified ranges: {}, {}, {}, {}",
rf.rannum_in(0.,100.), // wrapped f64 value
ru.rannum_in(1.,1000.), // wrapped u64, 1 to 1000 (inclusive)
ri.rannum_in(-10.,10.), // wrapped i64, -10 to 10 (inclusive)
ru16.rannum_in(60000.,65535.), // wrapped u16, 60000 to 65535 (inclusive)
ru8.rannum_in(1.,6.) // wrapped u8, 1 to 6 (inclusive)
);
They all print because Display has been implemented for Rnum.
The wrapped values can be if let pattern extracted as follows:
```rust use anyhow::{Result,bail};
if let Rnum::F64(x) = rf { utilise the x:f64 value }
else { bail!("rf does not hold value of f64 type!") };
``
The else branch can be used to report disappointed type expectations, as shown (assuming here thatanyhowcrate is being used for error handling). Alternatively,elsecan be used to return some default value, e.g.{0_f64}or it can be dismissed with a semicolon, usingif let` as a statement, rather than as an expression. In this case, should this particular extraction attempt fail, its associated action will be just quietly skipped:
rust
let uvec:Rv = ru8.ranv_in(20,1.,6.); // wrapped vec of random u8 values
if let Rv::U8(vx) = uvec {
println!("Dice roll sequence: {}", stringv(&vx)) };
This example illustrated the use of enum type Rv, used for returning whole vector of random numbers. As can be seen, its variants are extracted in the same way as from Rnum. Here helper function stringv (from module secondary.rs) converted the extracted vector to a String to facilitate its printing.
There is also enum type Rvv for returning vectors of vectors of random numbers:
rust
// vec of vecs using ranvv_in(d,n,min,max) and Display of Rvv
println!(
"5x5 matrix of integers in range [-10,10]:\n{}",
ri.ranvv_in(5,5,-10.,10.)
);
stringvv is another utility function to enable display of generic vectors of vectors. We did not need to use it here since Dislay is implemented for Rvv type and we did not bother to extract the wrapped value (vector of vectors).
The results wrapped within all three return types: Rnum,Rv,Rvv can all be pattern extracted as needed with if let.
Alternatively, for convenience, they can now all be extracted with supplied get functions, which just throw panic when the inner type is not found:
rust
// the following line tests 'getvi64()'
let pairofints = ri.ranv(2).getvi64();
println!("2 random i64s: {}", stringv(&pairofints));
Initialisation: the Self produced is Rnum type and will contain the default value zero of the required numeric end type.
rust
pub fn newf64() -> Self
pub fn newu64() -> Self
pub fn newi64() -> Self
pub fn newu16() -> Self
pub fn newu8() -> Self
The following methods are all implemented for Rnum, that means invoked on Rnum type variable. Even when generating Rv or Rvv type results. Rnum type input variable (self) in all cases serves just to inform the generic method about the numeric type required for the generated values:
pub fn rannum(&self) -> Self
returns a wrapped random number of one of the main types in maximum range allowed by the width of the type. The standardised range [0,1) is used for f64.
pub fn rannum_in(&self,min:f64,max:f64) -> Self
returns a wrapped random number of one of the main types in the range min,max (min,max are always f64s for commonality). The range should not exceed the width of the type, e.g. 0.,255. for u8. Nor should it be negative for unsigned types.
pub fn ranv(&self,d:usize) -> Rv
returns a wrapped Vec of length d filled with random numbers of one of the main primitive types. Note that the whole Vec is wrapped, not each individual element of it. Thus only one pattern extraction is needed.
pub fn ranv_in(&self,d:usize,min:f64,max:f64) -> Rv
same as ranv but using the specified range for the random values.
pub fn ranvv(&self,d:usize,n:usize) -> Rvv
returns a wrapped Vec<Vec<_>> consisting of n vectors, each of length d, filled with random numbers of one of the main primitive types. Note that only the whole result is wrapped, not each individual vector or element of it. Thus, again, only one pattern extraction is needed.
pub fn ranvv_in(&self,d:usize,n:usize,min:f64,max:f64) -> Rvv
same as ranvv but using the specified range for the random values.
There is no need to read beyond this point for normal daily use of this crate. However, there may be special circumstances, when using directly one of the typed functions is more convenient. Such as when needing only one specific end type. Another circumstance may be when wanting to use specific random number generator(s), different to the default ones used within the above methods. (Several are provided).
Utility functions to directly generate vectors of random numbers of common numeric end types:
```rust
/// Generates vector of size d,
/// filled with full range u64 random numbers.
pub fn ranvu64(d: usize) -> Vec
/// Generates vector of size d, of full range i64 random numbers.
pub fn ranvi64(d: usize) -> Vec
/// Generates vector of size d, of f64 random numbers in [0,1).
pub fn ranvf64(d: usize) -> Vec
/// Generates vector of size d, of u16 random numbers in [0,65535].
pub fn ranvu16(d: usize) -> Vec
/// Generates vector of size d, of u8 random numbers in [0,255].
pub fn ranvu8(d: usize) -> Vec
Utility functions to generate vectors of vectors (matrices) of random numbers of common numeric end types:
```rust
/// Generates n vectors of size d each,
/// filled with full range u64 random numbers.
pub fn ranvvu64(d: usize, n: usize) -> Vec
/// Generates n vectors of size d each, of full range i64 random numbers.
pub fn ranvvi64(d: usize, n: usize) -> Vec
/// Generates vector of size d, of i64 random numbers
/// in the interval [min,max]. May include zero.
pub fn ranvi64_in(d: usize, min:i64, max:i64) -> Vec
/// Generates n vectors of size d each, of f64 random numbers in [0,1).
pub fn ranvvf64(d: usize, n: usize) -> Vec
/// Generates n vectors of size d each, of u8 random numbers in [0,255].
pub fn ranvvu16(d: usize, n: usize) -> Vec /// Generates n vectors of size d each, of u8 random numbers in [0,255].
pub fn ranvvu8(d: usize, n: usize) -> Vec ``` And these f64 alternatives, using the improved f64 generator `` /// Generates n vectors of size d each, of f64 random numbers in [0,1).
pub fn ranvvf64_xoshi(d: usize, n: usize) -> Vec Some transformation wrappers for `` /// Generate u64 random number in the interval [min,max].
pub fn ran_urange(min:u64, max:u64) -> u64 /// Generate i64 random number in the interval [min,max].
pub fn ran_irange(min:i64, max:i64) -> i64
``` There is also a function that transforms any f64 number in standard range [0,1) to a new range: Version 0.3.3 Some reorganisation. Added module Version 0.3.2 Added U16 type random numbers generation. Version 0.3.1 Updated README.md to read more like an introductory user manual. Version 0.3.0 Substantial revision. Completed the generic interface. Renamed some types and functions. Created two separate source modules ( Version 0.2.4 Extended the interface to vecs and vecs of vecs, in full range and in given range. Version 0.2.3 Added boilerplate polymorphic interface. Version 0.2.2 Added Version 0.2.0 Added Version 0.1.4 Fixed the debug mode overflow 'errors'. They were not affecting the release mode but given that this crate is intended for testing, they were annoying. The solution was to use Version 0.1.3 Tested and fixed an &mut argument. Added Version 0.1.2 Fixed the initial typos. Version 0.1.1 Changed the crate name to Version 0.1.0 The initial version.
xoshif64():rust
/// Generates vector of size d, of f64 random numbers in [0,1).
/// Bit slower but otherwise superior toranvf64`.
pub fn ranvf64_xoshi(d: usize) -> VecLow Level Integer Algorithms
xoshiu64() generates u64 random numbers in full 64 bit range and 2^256 state space. That means the sequence is not going to repeat for a very long time. This algorithm is used to construct random numbers of all (unsigned) integer types and ranges up to 64 bits.splitmix() also generates u64 numbers. It is used here only to generate the initial seeds for the 'xoshi' type algorithms.xoshiu64():rust
/// Get random numbers of various smaller unsigned integer
/// types, by specifying the number of bits required,
/// e.g.ranubits(16) as u16`
pub fn ranubits(bits:u8) -> u64 Low Level Floating Point Algorithms
ranf64() is a little older (George Marsaglia, 2003). It has been adapted here to generate f64 numbers in the standard range: half open interval [0,1). That means its output can be easily transformed into any other range. Its main claim to fame is its superior speed.xoshif64() is also fast, though not quite as much as ranf64() but it makes up for it by quality. It has also been adapted to output f64 numbers in the standard range [0,1).rust
/// Transform f64 number in [0,1) to [min,max)
pub fn ran_ftrans(rnum:f64, min:f64, max:f64) -> f64
Release Notes (Latest First)
generators.rs which now contains all generating code. Added get functions to impls.rs for easy extraction of the inner values of all supported types from Rnum, Rv, Rvv.impls.rs, secondary.rs) to clean up lib.rs. Removed dev-dependence on indxvec.ran_irange, ranvi64, ranvvi64, to generate i64 random numbers in any i64 range. Plus some appropriate tests in tests.rs. Restricted bits argument in ran_ubits(bits:u8) to u8, as it should never exceed even 63. Corrected some comments.tests/tests.rs. Added general purpose xoshiu64() which is now used to construct random numbers of all (unsigned) integer types and ranges. Reorganised, renamed and/or deleted some functions. Made the xoshi seeds also static, for ease of use. They no longer need to be explicitly passed as arguments.overflowing_ versions of some of the operators, available as of Rust version 1.53. So, in case of problems, you may have to update to the latest (stable) version of Rust.ranvvf64_xoshiro and ranvvu8_xoshiro for completeness.ran as all others are taken.