Include sympy.stats?

[toomey8]

Would it be possible to extend support to cover https://docs.sympy.org/latest/modules/stats.html/?

SymPy.jl is fantastic! Many thanks!

[jverzani]

You can already access it with a command like:

julia> SymPy.PyCall.pyimport_conda("sympy.stats", "sympy")

Then you can access the functions with the sympy.stats prefix:

julia> Z = sympy.stats.Normal("Z", 0, 2)
Z

julia> sympy.stats.E(Z)
0

If you wanted to make PR where a module gives more direct access the function names, that would be helpful.
However, if you wanted to import all the symbols, you could simply do:

SymPy.import_from(sympy.stats)

Then you could do things from the documentation, though still with minor modification. E.g.:

julia> p = 1//2
1//2

julia> @vars x integer=true positive=true
(x,)

julia> pdf = p * (1-p)^(x-1)
 1 - x
2     
──────
  2   

julia> D = DiscreteRV(x, pdf, set=sympy.S.Naturals)
x

julia> E(D)
2

julia> P(D ≫ 3)
1/8

[toomey8]

Many thanks this is perfect! It's signifigantly above my Julia grade to submit such a PR, but for the archives might note that the below worked perfectly for me.

I was mucked up on the syntax on how to call the SymPy module.

using SymPy, PyCall;
SymPy.PyCall.pyimport_conda("sympy.stats", "sympy")
E(x) = sympy.stats.E((x))
sample(x) = sympy.stats.sample(x)
Normal(x,y,z) = sympy.stats.Normal(x,y,z)
Z = Normal("Z", 0, 2)
E(Z)

Again, sincere thanks!

[newptcai]

Any plan to add this in SymPy.jl by default?

[jverzani]

It depends on what you mean by this.

We could include just the lines

using SymPy, PyCall;
SymPy.PyCall.pyimport_conda("sympy.stats", "sympy")

easily enough.

The import_from could be added to automatically export the methods. But I wouldn't make that a default as it enlarges the namespace too much. We could make a module in SymPy that would basically include these lines:

module Stats

SymPy.PyCall.pyimport_conda("sympy.stats", "sympy")
import_from(sympy.stats)
end

Then users could opt in if they chose through using SymPy.Stats. Adding documentation would be nice as some things don't quite work (e.g. if Z is an RV, the Z>6 isn't going to work, as in SymPy.jl that returns a boolean, but Gt(Z,6) does.

If you wanted to make a PR, I'd be quite willing to merge something like that in.