I wanna fill my array in Julia by positive real numbers. But I found information only how to do it with integers or real numbers (including negatives). Is it possible? Thanks!
2 Answers
You can use any mathematical formula that maps the [0, 1) range into the [0, +inf]:
For example, if x is your random variable in the [0, 1) range (obtained with e.g. rand() for float data types):
tan(x * pi / 2)atanh(x)log(x) ^ 2-log(x) / x ^ p(forpnon-negative integer -- it will change the number distribution)
There are many other functions.
Of course the numbers are no longer uniformly distributed, but that is impossible to achieve.
Technically, the built-in randexp fulfils your requirement: the exponential distribution has the positive reals as its support. The scale of the numbers you practically get is much lower, though. The same holds for abs ∘ randn, the half-normal distribution. (In both cases, you could multiply the results with a large positive number to increase the variance to your requirements.)
Here's a funny alternative: you can generate uniformly random bits, and reinterpret them as floats (and just set the sign always to positive):
julia> bitrand(3*64).chunks
3-element Vector{UInt64}:
0xe7c7c52703987e68
0xc221b9864e7bab7e
0xa45b39faa65b446e
julia> reinterpret(Float64, bitrand(3*64).chunks)
3-element reinterpret(Float64, ::Vector{UInt64}):
2.8135484124856866e-108
-4.521596431965459e53
-5.836451011310255e78
julia> abs.(reinterpret(Float64, bitrand(3*64).chunks))
3-element Vector{Float64}:
1.6467305137006711e236
3.3503597018864875e-260
1.2211675821672628e77
julia> bitstring.(abs.(reinterpret(Float64, bitrand(3*64).chunks)))
3-element Vector{String}:
"0000110000011000001110000110001111010000011110111101000101101101"
"0011000010110101111100111011110100111100011011000101001100010011"
"0110111000001000101011010100011011010010100111111011001000001100"
This is still not a uniform distribution on the values, though, as the precision of floats gets smaller the larger the exponent gets.
Related