[ANN] Telperion.jl - Macro Magic for Properties and Statistical Formulas

Package Announcements
,
1

Hi everyone,

I created a package to help me work with properties in an expression without repeatedly using the dot syntax. The gist is that you can do this:

julia> nt = (x=1, y=2)
(x = 1, y = 2)

julia> z = 3
3

julia> @withprops nt x/y + z
3.5

That is, for the statement @withprops src expr, any symbol x that shows up in expr will be replaced by getproperty(src, x). This likely already exists somewhere else, but I couldn’t find it.

My main use of this was creating my own StatsModels-like syntax, where instead of creating columns via a DSL, each term is valid Julia code (after the getproperty replacement) e.g.

using DataFrames, StatsBase, Telperion

df = DataFrame(y=rand(100), a=1:100, b=randn(100), c=randn(100), d=rand(1:5, 100))

x, y = @xy df log.(y) ~ 1 + a + zscore(b) + abs.(sin.(c)) + dummy(d)

x

OrderedDict{String,Any} with 8 entries:
  "1"             => [1, 1, 1, 1, 1, 1, 1, 1, 1, 1  …  1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
  "a"             => [1, 2, 3, 4, 5, 6, 7, 8, 9, 10  …  91, 92, 93, 94, 95, 96, 97, 98, 99, 100]
  "zscore(b)"     => [1.13036, -0.280105, 2.29973, -0.267989, -0.240071, -0.797709, -0.315514, -0.322103, 0.0217353, -1.67589  …  1.45323, -0.363556, -0.650576, -1.543…
  "abs.(sin.(c))" => [0.753822, 0.992965, 0.41306, 0.733578, 0.21487, 0.958583, 0.163681, 0.238074, 0.166078, 0.920199  …  0.407876, 0.277916, 0.0207317, 0.572013, 0.2…
  "dummy(d) [2]"  => Bool[0, 0, 0, 0, 0, 0, 1, 0, 0, 0  …  0, 1, 0, 0, 0, 0, 1, 0, 0, 0]
  "dummy(d) [3]"  => Bool[0, 0, 0, 0, 0, 1, 0, 1, 0, 1  …  0, 0, 0, 0, 0, 1, 0, 0, 0, 0]
  "dummy(d) [4]"  => Bool[0, 1, 0, 1, 0, 0, 0, 0, 0, 0  …  0, 0, 1, 0, 1, 0, 0, 0, 0, 0]
  "dummy(d) [5]"  => Bool[1, 0, 1, 0, 1, 0, 0, 0, 0, 0  …  1, 0, 0, 1, 0, 0, 0, 0, 0, 1]

The idea is that you can then create the matrix of features via reduce(hcat, values(x)) or similar.

Here’s the source: GitHub - joshday/Telperion.jl: Simple Statistical Formulas

p.s. There’s nothing to the name. It’s just a random Tolkien reference.

3 Likes
2

This looks promising. Thanks!

3

@Mason has something similar to @with in StatidModules.jl. It works in I think the same way.

Another note is that in general, when you use @withprops with DataFrames, Julia won’t be able to write very performant code, since Julia doesn’t know from the type of df which symbols it encounters are in the data frame.

4

Thanks for the link! I didn’t know about StaticModules. I like that you can use multiple objects in the @with blocks.

I borrowed that idea in https://github.com/joshday/PropertyUtils.jl (just put together this morning) but decided to use a function so that you can compose different property-changing functions together.

using PropertyUtils

a = (x = 1, y = 2)
b = (y = 3, z = 4)

struct A 
    x::Int 
end
Base.getproperty(::A, x::Symbol) = "hello!"

result = @with joinprops(fields(A(10)), a, b) begin 
    x + y
end

result == 12