Implement S[I::AbstractVector{Bool}] and S[I::AbstractVector] for sparse X

[ViralBShah]

julia> s = speye(5); x =  [true,false,true];

julia> s[x]
ERROR: no method similar(SparseMatrixCSC{Float64,Int64}, Type{Float64}, (Int64,))
 in getindex at abstractarray.jl:363

julia> s = speye(5); y =  [1,2,3];

julia> s[y]
ERROR: no method similar(SparseMatrixCSC{Float64,Int64}, Type{Float64}, (Int64,))
 in getindex at abstractarray.jl:363

Cc: @tanmaykm

[ViralBShah]

The mention of this issue in #7006 was done mistakenly and is unrelated.

[tanmaykm]

Indexing with boolean array fails unless it is of same size as the array. Is this what is desired? This is not the behavior in octave and matlab.

julia> a = rand(5, 5)
5x5 Array{Float64,2}:
 0.904719   0.659319  0.108763  0.439896  0.769862
 0.0666711  0.860941  0.830392  0.634226  0.423541
 0.940728   0.681775  0.345935  0.253252  0.673635
 0.147969   0.850448  0.57376   0.2739    0.97953 
 0.303279   0.970842  0.382316  0.954433  0.57161 

julia> a[[true,false,true]]
ERROR: BoundsError()
 in getindex_bool_1d at array.jl:285
 in getindex at array.jl:300

[nalimilan]

I think this is intentional and related to the "no recycling" rule. It could be added to the Noteworthy differences section of the manual for Matlab. For now it's explained in the R section in PR #6221:

+- Like most languages, Julia does not, by default, allow operations on unequal
+  length vectors by cycling the shorter one. Julia will throw errors when the
+  lengths of inputs do not match, unlike R in which c(1,2,3,4) + c(1,2) will
+  evaluate to c(2,4,4,6).

[pao]

...I'm not even sure what MATLAB's rule is here. Seems better to avoid this entirely.

>> a = magic(5)
a =
    17    24     1     8    15
    23     5     7    14    16
     4     6    13    20    22
    10    12    19    21     3
    11    18    25     2     9
>> a(logical([0 1; 0 1]))
ans =
     4
    10

[tkelman]

Matlab's rule appears to be a(b(:)) for a(b), and if b is shorter than a then b is effectively padded with false's. +1 for just erroring though (perhaps a nicer more specific error), that cycling behavior from R makes me throw up a little.

[pao]

@tkelman ahh!

I think broadcast_getindex should probably have us covered.

[tanmaykm]

Thanks. I think that's fair. Probably just needs a better mention in the array indexing section of the manual.

[jiahao]

Matlab's behavior in this example is consistent with the general indexing rule that a smaller array behaves as if it is zero-padded beyond its actual dimensions, and is allowed to grow by zero-padding if necessary. This specific case of logical indexing is documented here in the subsection "Logical Indexing with a Smaller Array".

[pao]

Matlab's behavior in this example is consistent with the general indexing rule that a smaller array behaves as if it is zero-padded beyond its actual dimensions, and is allowed to grow by zero-padding if necessary.

But I was not expecting MATLAB to columnize the matrix first.

[jiahao]

But I was not expecting MATLAB to columnize the matrix first.

Matlab calls this linear indexing. It's not documented, but I believe Matlab resorts to linear indexing whenever ordinary indexing doesn't make sense. (Implicitly flattening the matrix is part of the attempt to apply linear indexing semantics.) The interaction of all of Matlab's indexing rules can be rather intricate to figure out.

[pao]

I know it's linear indexing. I've just never seen it do that with a logical index! (Or if I have, I filed it into the "don't ever do that again" category).

At any rate, we seem to all agree it's not something we should emulate. Consider this horse thoroughly flogged.

[jiahao]

I know you know. It was more a comment for posterity. ;)

[ViralBShah]

#7047 closes this. The behaviour for logical indexing is the same as the dense case.