How to get the average of a subset of data in a matrix?

I am having some difficulty finding the average of some subsets of data. For example, given the following matrix:

ItemA 7

ItemA 8

ItemA 9

ItemB 10

ItemB 12

ItemB 14

ItemC 8

ItemC 16

ItemC 0

Given the above data, I would like to find the average of item A, B, and C individually. So it should return that Item A the average is 8, Item B the average is 12, and item C the average is 8.

The above data is just an example, the actual data is many more items so I need it to work for an unknown amount of items.

If you have any questions feel free to ask,

Thanks

[Image Analyst]

What form is the data in? Is it a N by 2 cell array, where the first column is the itemX word and the second column is the number?

[Aaron]

I think I tried to simplify the data too much. A better representation of my data would be the following:

Item Xcoord YCoord

sample data could be:

ItemA 1 3

ItemA 2 5

ItemA 3 2

ItemB 2 4

ItemB 1 3

ItemC 2 9

ItemC 9 2

Given the above data I am actually wanting to run kmeans on item A, kmeans on item B, and kmeans on item C, but for this question I am mainly just wanting to know how I can run an operation such as averaging, kmeans, etc on each item's data individually. The data is in the format Item# first column, Xcoord second column, Ycoord third column.

[Aaron]

And it is an Nx3 cell array, but for the kmeans it takes in a matrix so I would need to have columns 2 and 3 converted to a matrix before k means can use it

[Image Analyst]

I don't understand how kmeans comes into this. For each number in either column 2 or column 3 you know the category (A, B, or C) it falls into. So what is there to do with kmeans? How can you cluster/classify this data when it already is clustered/categorized? You can do the mean of each category - no problem with that - but kmeans???? I don't know how or why that would apply.

[Aaron]

I run kmeans and say that there is one cluster. Is that the same thing as just taking the mean of all the x's and the mean of all the y's? I thought that the advantage of using kmeans,1 would be that it would be weighted such that the outliers would not count as much in the average.

[Aaron]

If the mean of the x's and the mean of the y's are the same as kmeans of one cluster then I would be able to just use accummarray right?

[Aaron]

Also, I think kmeans tries to minimize the Euclidean distances to give a centroid with the minimal distance between all of the data points. I am open for suggestions though

Answer 1

[Andrei Bobrov]

C = {'ItemA' 1 3
'ItemA' 2 5
'ItemA' 3 2
'ItemB' 2 4
'ItemB' 1 3
'ItemC' 2 9
'ItemC' 9 2};

[a,ii,ii] = unique(C(:,1));
[j1,j2] = ndgrid(ii,1:2);
b = cat(1,C{:,2:3});
out = [a, accumarray([j1(:),j2(:)],b,[],@(x){mean(x)})];

[Aaron]

That worked out perfectly. Is there a way to do standard deviation in the accumarray function instead of mean? like @stddev or something similar. Thanks a lot for your answer, it was very useful.