How to find rank within group

I have a dataset with 100 different groups, and between 8-11 entries in each group. I want to rank them, within a group, by the value of a variable y. How would I do this?

Concretely, if I have 2 groups with 2 people each such that: y=[1; 3; 6; 5] group_id=[1;1;2;2]

then I want a third vector: rank=[1;2;2;1]

Answer 1

[Andrei Bobrov]

[~,a] = cellfun(@sort,accumarray(group_id,y,[],@(x){x}),'un',0);
rank = cell2mat(a);

or within for..end:

y=[1; 3; 6; 5];
group_id=[1;1;2;2];

rank = group_id;
a = unique(group_id);
for ii = 1:numel(a)
    t = group_id == a(ii);
    [~,rank(t)] = sort(y(t));
end

[Garima Sharma]

Thanks, this works!

[Simon]

The result created by the cellfun-based method is different from the for-loop based method when group_id is not in sorted order. The within-group ranking is the same in both methods. But rank = cell2mat(a) put all 1st group's within-group ranks together, followed by the second group's.

I have came across similar problem and in a subsequent analysis I need to put rank as a new column side-by-side next to group_id. The cellfun-based method would give me erroneous result.

Answer 2

[Jess Lovering]

Do you mean that you want to sort the data by a specific array? You can do this using the sort function. It will also give you the indices of the new order so you can use that to apply to the other arrays, example:

y=[1; 3; 6; 5] 
group_id=[1;1;2;2]
rank=[1;2;2;1]
[sorted_y,I]= sort(y)
sorted_group_id = group_id(I)
sorted_rank = rank(I)

[Garima Sharma]

no, what I want to do is rank y=[1; 3; 6; 5] within its respective group_id=[1;1;2;2].

So, y(1,1) is rank 1 in group 1; y(2,1) is rank 2 in group 1. y(3,1) is rank 2 in group 2, and y(4,1) is rank 1 in group 2. I want to only input vectors y and group, and get vector rank as the output.