Octave group statistic computation with accumarray and user defined function returning a third column's value - arrays

Question

To be clear the following is not my original problem which has data that is much larger and this code is in the context of a larger application and code base. I have reduced my work to the simplest example that’s now at toy or didactic size for clarity and dev and unit testing because that helps a lot for these purposes as well as for sharing on stackexchange. I am experienced in R but not in octave (Matlab). This is code for octave version 4.0.0. I seem to be stuck on translating group computations such as R’s tapply() or by() as well as writing and calling user defined functions (plus a bit of additional processing than those built-ins), but now written in the octave language.
Starting state is an array a as shown:
a = [5 1 8 0; 2 1 9 0; 2 3 3 0; 5 3 9 0]
a =
5 1 8 0
2 1 9 0
2 3 3 0
5 3 9 0
The process I need to do is essentially just this: Group by column 1, find the min statistic in column 3, return the value stored in column 2 of the same row, and write the value to column 4. I want no optional packages to be used. The built-in accumarray and min functions together get me pretty close but I’ve not found the needed syntax. Matlab seems to have many versions of parameter passing syntaxes developed over different releases and please note my code needs to run in Octave 4.0.0.
Final state desired is same array a, but column 4 is updated as shown:
a =
5 1 8 1
2 1 9 3
2 3 3 3
5 3 9 1
My best few code snippets of near-misses and most interesting things among all my failed attempts (not shown, as there are many pages of attempts that do not work) are:
[x,y] = min(a(a(:,1)==5,3),[],1)
x = 8
y = 1
Notice that y is index of row within the group, but not row within the a array, which is fine and good as long as I later do a computation to translate indexes from group-relative to global-relative, and inside there read the value of a(y,2) which is the correct answer value for each row.
>> [x,y] = min(a(a(:,1)==2,3),[],1)
x = 3
y = 2
>> [~,y] = min(a(a(:,1)==2,3),[],1);
>> y
y = 2
Notice that y is all I need from min() since it’s the index of the row of interest.
>> accumarray(a(:,1), a(:,3), [], #([~,y]=min([],[],1)))
parse error:
syntax error
Notice that with some kind of syntax I need to pass to min() in its first parameter the group of values determined by parameters 1 and 2 of accumarray.
I ultimately need to have something like this happen within the group computations after min() returns row index y:
a(y,4) = a(y,2); % y is the desired row index found by min() within each group
So, I tried to write a function that’s named for possibly simpler syntax:
>> function idx = ccen(d)
[~,y]=min(d,[],1);
idx=a(y,2);
end
>> accumarray(a(:,1), a(:,3), [], #ccen)
error: 'a' undefined near line 3 column 5
error: called from
ccen at line 3 column 4
accumarray at line 345 column 14
Seems to me, that to my surprise, a is not accessible to function ccen. Now what can I do? Thank you for reading.

Answer 1

When declaring functions in MATLAB / Octave, any variables declared outside the scope (by default) are not accessible. This means that even though you have a declaration for a, when you create that function, a is not accessible within the scope of the function.
What you can do is modify ccen so that a is supplied to the function so it can access the variable when the function is being called. After, wrap an anonymous function around your call to ccen when calling accumarray. Anonymous functions however do have the luxury of capturing the scope of variables that aren't explicitly declared as input variables into the function:
So first:
function idx = ccen(a, d) %// Change
[~,y]=min(d,[],1);
idx=a(y,2);
end
And now...
out = accumarray(a(:,1), a(:,3), [], #(x) ccen(a,x)); %// Change last parameter
This call is acceptable because the anonymous function is capturing a at the time of creation. Notice how x in the anonymous function is what is piped in from the accumarray calls. You're simply forwarding that as the second parameter to ccen and keeping a constant. This doesn't change the way the function is being run.... it's just resolving a scope issue.
I get the following in Octave:
octave:10> a = [5 1 8 0; 2 1 9 0; 2 3 3 0; 5 3 9 0]
a =
5 1 8 0
2 1 9 0
2 3 3 0
5 3 9 0
octave:11> function idx = ccen(a,d)
> [~,y]=min(d,[],1);
> idx=a(y,2);
> end
octave:12> out = accumarray(a(:,1), a(:,3), [], #(x) ccen(a,x))
out =
0
1
0
0
1