countmap weird behaviour for -0.0

[jhbdiehl]

t = [0., -0.]
ct = countmap(t, [1 for e in t])

will output:

Dict{Float64, Int64} with 2 entries:
  0.0  => 1
  -0.0 => 1

Changing t to Integers

t = [0, -0]
ct = countmap(t, [1 for e in t])

or removing the weights

t = [0., -0.]
ct = countmap(t)

leads to the expected behaviour:

Dict{Float64, Int64} with 1 entry:
  0.0 => 2

Now all of this does not seem problematic until you start changing the keys, e.g.

Dict(keys(ct) .+ 1 .=> values(ct))

leads to

Dict{Float64, Int64} with 1 entry:
  1.0 => 1

in the first case, instead of

Dict{Float64, Int64} with 1 entry:
  1.0 => 2

Therefore I think this is not the intended behaviour.
To be fair this could also be interpreted as a bug in the general Dict routine.

[nalimilan]

Good catch. See #831. This may end up going in the opposite direction you expected though.

Note that the different behavior of t = [0, -0] isn't a bug in StatsBase, it's just that -0 it exactly the same as 0, as the equivalent of -0.0 doesn't exist for Int. In other words, the information is lost even before countmap has the chance to see it.

Likewise, Dict(keys(ct) .+ 1 .=> values(ct)) simply retains the value for the last time the key was seen in (keys(ct) .+ 1 .=> values(ct)) == [1.0 => 1, 1.0 => 1], it doesn't sum counts (as the Dict constructor isn't made to work only with counts but with any kind of data).