convert matrix to tensor

i have a data set which has the following structure road_segment_id/<time_slot>/<driver_id>/<travel_time> Now I need to select randomly 100 driver ids and create a 3-d tensor with road_segment_id,driver_id,time-slot on 3 sides.Let the tensor be A then a(i,j,k)=c represents that time taken by driver j to traverse road segment i on time-slot k is c. the attached file showing a part of data set is there.

[SOUMYA SUVRA GHOSAL]

here's my code,please tell me where is the error

 if 
road=unique(M(:,1));
time_slot=unique(M(:,2));
user_id=unique(M(:,3));
n_road=numel(road);
n_time=numel(time_slot);
n_user=numel(user_id);
driver=intersect(user_id,randsample(user_id,100));
A=ones(n_road,100,n_time);
for ii=1:n_road
    for jj=1:100
        for kk=1:n_time
            [r]=find(M(:,1)==road(ii) & M(:,3)==driver(jj) & M(:,2)==time(kk));
            A(ii,jj,kk)=M(r,4);
        end
    end
end

 A is the 3d array 
its giving error that
"Assignment has more non-singleton rhs dimensions than non-singleton subscripts"

Error in tensor A(ii,jj,kk)=M(r,4);

Answer 1

[Aakash Deep]

I think this will help you. You can refer this https://www.mathworks.com/matlabcentral/answers/27338-convert-matrix-into-the-tensor.

If this doesn't work you can still ask about it but try to mention more information about the dimensions of input matrix and output tensor.

[SOUMYA SUVRA GHOSAL]

My matrix is of dimension 4432506×4 and my tensor is a 3 dimensional tensor of 99320 * 100 *8 I tried applying the same procedure you mentioned but it is a cell array and I need a tensor. From the unique command I could find out the number of road segments,drivers and taxis but problem is how to enter the value of travel time at each point.

[Aakash Deep]

Hey Soumya, yes you are right that the error is in below statement

A(ii,jj,kk)=M(r,4);

because when you execute below statement

[r]=find(M(:,1)==road(ii) & M(:,3)==driver(jj) & M(:,2)==time(kk));

there might be chances that we will get a vector because there are possibilities that more than 1 index values are same. So, this will give a vector of nx1 dimension. Now, you are trying to insert this vector into a cell of a tensor i.e. A(ii,jj,kk)

Hope this helps :)

[SOUMYA SUVRA GHOSAL]

Thank you so much Aakash, can you do me one last favour by telling how to modify the code. Means what should I write in the assignment statement.

[Aakash Deep]

Hey Soumya, your approach is correct to populate the tensor. I am not exactly able to figure out why it is behaving like this but here are the few tips you should try:

• After looking at the below statement,

[r]=find(M(:,1)==road(ii) & M(:,3)==driver(jj) & M(:,2)==time(kk));

I am assuming that the error is coming from M(:,3)==driver(jj) this part because the value of road can repeat as many time it wants and the value of time can also repeat but what can not repeat is the driver_id for the same pair of road and time (if it does then there is redundancy is your data).

• There can be a possibility that the vector r contains same value multiple times. So just try to use unique on r and then use it as an index value.
• There are highly chances that the vector r can be a null set because of the & operator so you should first check it if it is a single value or not before using it as a index value during assignment. This will fix some of your errors.

Hope this helps :)