I am having trouble understanding what seems like an inconsistent behavior in Julia.
X = reshape(1:100, 10, 10)
b = [1 5 9]
X[2, :][b] # returns the correct array
X[2, :][1 5 9] # throws an error
Can someone explain why using the variable b works to index an array but not when I write the index myself?
Since x = X[2,:] is just a vector, we can simplify the example to just talking about indexing behavior on vectors.
x[v] where v is a collection of integers returns the subset of x. Thus x[(1,5,9)], or x[[1,5,9]] is thus using that getindex(x::Vector,i::AbstractArray) dispatch.
Note that x[[1 5 9]] works because v = [1 5 9] makes v a row vector. That's valid syntax, but x[1 5 9] just isn't even valid Julia syntax. That syntax means something else:
v = Float64[1 5 9]
returns a row vector with element type Float64.
X = Float64; X[1 5 9]. And error is not from parsing but just about typed_hcat(T, ...) missing. But I am not able to define appropriate typed_hcat function! Why?
typed_hcat is for what I showed up there: the first argument should be a data type that is the element-type for the resulting array. I believe this is built into the core Julia since it's special parsing and handling for array construction.
Base.typed_hcat(::AbstractArray, ::Int...) = "Don't do this!".
(1:10)[[1 4]] and (1:10)[[1, 4]].' result in different types. I wonder if there's a valid reason for this, or just a historical remnant of how the two were developed so far.(1:10)[[1, 4]]. Try this (1:10)[[1, 4],:] or this (1:10)[[1, 4],:].' (or reshape((1:10)[[1, 4]], (2,1)) )I have figured out a solution.
Rather than write X[2, :][1 5 9] I should have written x[2, :][[1 5 9]]
I believe this makes sense when we imagine indexing on two dimensions the second time. This makes it possible to write more complicate indices, like X[2:4, :][[1 3],[1 3]]
X[2:4, :][[1 3],[1 3]] returns 4 elements rather than 2 (which might be what you expect if you're coming from python).