unique function Matlab<->Julia

[Stephan Buchert]

In Matlab I can do, for example, with "daynum" fractional day numbers:

idaynum = floor(daynum);
[umjd, im, iu] = unique(idaynum);  % <= how do I get the "iu" in Julia?

% I need an observation "pms" listed per day in a file.
% For performance reasons the parameter is retrieved from the file only once for each of the different integer days
upms = pi*umjd;   % <= for illustration only, instead of really retrieving the observed parameter
pms = upms(iu);
% Now "pms" matches "daynum", ie. when daynum=[1.0, 1.5, 1.8, 2.0, 2.2, 3.1], then pms approx. [3.14, 3.14, 3.14, 6.28, 6.28, 9.42]

Probably this can be done elegantly also in Julia, but, coming from Matlab, for me not easy to see how?

[Tomas Lycken]

For completeness, here is an excerpt from the MATLAB help file on `unique`:

[C,ia,ic] = unique(A) also returns index vectors ia and ic.

• If A is a vector, then C = A(ia) and A = C(ic).

• If A is a matrix or array, then C = A(ia) and A(:) = C(ic).

• If A is a table, then C = A(ia,:) and A = C(ic,:).

Which of the cases is relevant for you?

// T

[Stephan Buchert]

The case "A is a vector" is relevant for this, and I need "ic", for A=C(ic)

[Steven G. Johnson]

On Tuesday, April 21, 2015 at 5:16:12 AM UTC-4, Stephan Buchert wrote:

In Matlab I can do, for example, with "daynum" fractional day numbers:

idaynum = floor(daynum);
[umjd, im, iu] = unique(idaynum);  % <= how do I get the "iu" in Julia?

This is not implemented yet; see:

https://github.com/JuliaLang/julia/issues/1845

[Stephan Buchert]

Ok, thanks. At last here is a use case.

[Tim Holy]

Interestingly, even though I use(d) the 3-output `unique` in Matlab on an
almost daily basis, in Julia I never find I need/want it---I didn't even
realize we didn't have it. I confess to not being sure what the difference is,
but I suspect it's because there must be better approaches in julia for the
problems for which I would have used `unique`.

--Tim

[Dawid Crivelli]

You can probably get around it, by using the `indexin` function

redunant = rand(1:4,10)    # [1,3,3,4,3,1,3,3,3,4]
slimmed = unique(redunant) # returns in my case [1,2,3,4]

indexin(slimmed, redunant) # returns [6, 9, 10], which is one of the possibilities

It's almost easier to modify either `unique` or modify one of the `findin` or `indexin` functions to make your own unique which returns indexes