BUG: funeval returns a 0 dimensional array when BasisStructure has single point

[sglyon]

Issue by spencerlyon2
Friday Jul 31, 2015 at 02:16 GMT
Originally opened as QuantEcon/CompEcon.jl#10

---

Test case:

using CompEcon
basis = Basis(ChebParams(10, -1, 1), ChebParams(5, 2, 4));
grid = nodes(basis)[1];
y = grid[:, 1] - grid[:, 2];  # simple function x - y
c = funfitxy(basis, grid, y)[1];
pt = [0.5, 2.1];
bs = BasisStructure(basis, Direct(), pt);
funeval(c, bs)

REPL example:

julia> using CompEcon

julia> basis = Basis(ChebParams(10, -1, 1), ChebParams(5, 2, 4));


julia> grid = nodes(basis)[1];

julia> y = grid[:, 1] - grid[:, 2];  # simple function x - y

julia> c = funfitxy(basis, grid, y)[1];

julia> pt = [0.5, 2.1];

julia> bs = BasisStructure(basis, Direct(), pt);

julia> funeval(c, bs)
0-dimensional Array{Float64,0}:
-1.6

[sglyon]

Comment by spencerlyon2
Friday Jul 31, 2015 at 02:19 GMT

---

Could be fixed by changing

function squeeze_trail(x::Array)
    sz = size(x)
    squeezers = Int[]
    n = length(sz)
    for i=n:-1:1
        if sz[i] == 1
            push!(squeezers, i)
        else
            break
        end
    end
    squeeze(x, tuple(squeezers...))
end

to

function squeeze_trail(x::Array)
    sz = size(x)
    squeezers = Int[]
    n = length(sz)
    for i=n:-1:1
        if sz[i] == 1
            push!(squeezers, i)
        else
            break
        end
    end
    out = squeeze(x, tuple(squeezers...))
    length(out) == 1 ? out[1] : out
end

but I think with some design work we can do a better job at finding a type-stable version of funeval (because as of right now it is very type unstable...)

[sglyon]

Comment by spencerlyon2
Friday Jul 31, 2015 at 02:21 GMT

---

See type instability by continuing previous example and doing @code_warntype funeval(c, bs, fill(0, 1, 1)) (if you are on julia 0.4dev)