Lookup values in other matrix

I have two very large vectors of the same size:

Vector 1 contains a list of type

Type ("Math","Chem","Bio")
["Math" "Chem" "Bio" "Chem" "Chem" "Math" ...]

Vector 2 contains a different list of types

Type ("Baseball","Tennis","Soccer")
["Baseball" "Tennis" "Soccer" "Tennis" "Tennis" "Baseball" ...]

I have another small matrix with numeric values associated with each combination of the two types:

"Math" "Baseball" 5
"Math" "Tennis"   8
"Chem" "Baseball" 16
.....

I would like to create a new vector to lookup base on the matrix for the numeric value. What would be the vectorized way to deal with this situation? Seems like a very common task, but I couldn't figure it out.

Thanks in advance, Fischer

Answer 1

[dpb]

>> v1={'Math'; 'Chem'; 'Bio'; 'Chem'; 'Chem'; 'Math'};
>> v2={'Baseball' 'Tennis' 'Soccer' 'Tennis' 'Tennis' 'Baseball'}.';
>> A={'Math' 'Baseball' 5;
'Math' 'Tennis'   8;
'Chem' 'Baseball' 16};
>> A(ismember([char(A(:,1)) char(A(:,2))], ...
              [char(v1) char(v2)],'rows'),:)
ans = 
    'Math'    'Baseball'    [5]
>>

ADDENDUM Per comment below, reorder; use alternate return index--

>> [~,ib]=ismember([char(v1) char(v2)], ...
                   [char(A(:,1)) char(A(:,2))],'rows')
ib =
     1        0       0       0       0       1
>> for i=1:length(ib)
     if ib(i)
       A(ib(i),:)
     else
       disp('?')
     end
   end
ans = 
    'Math'    'Baseball'    [5]
?
?
?
?
ans = 
    'Math'    'Baseball'    [5]
>>

ADDENDUM 2 To clarify above remark, built full table with arbitrary (but satisfy the previous partial values) assignments for the numerical values as follows--

>> u1=unique(v1);  % unique subjects
u1 = 
  'Bio'
  'Chem'
  'Math'
>> a=[6 12 1]; % corresponding values
>> u2=unique(v2) % ditto for sports
u2 = 
  'Baseball'
  'Soccer'
  'Tennis'
>> s=[4 6 7];
>> k=0;  % now build the table
 for i=1:3,for j=1:3,
   k=k+1;
   A(k,:)={u1{i} u2{j} a(i)+s(j)};
 end,end
>> A
A = 
  'Bio'     'Baseball'    [10]
  'Bio'     'Soccer'      [12]
  'Bio'     'Tennis'      [13]
  'Chem'    'Baseball'    [16]
  'Chem'    'Soccer'      [18]
  'Chem'    'Tennis'      [19]
  'Math'    'Baseball'    [ 5]
  'Math'    'Soccer'      [ 7]
  'Math'    'Tennis'      [ 8]
>> [~,ib]=ismember([char(v1) char(v2)], ...
                 [char(A(:,1)) char(A(:,2))],'rows');
>> A(ib,:)
ans = 
  'Math'    'Baseball'    [ 5]
  'Chem'    'Tennis'      [19]
  'Bio'     'Soccer'      [12]
  'Chem'    'Tennis'      [19]
  'Chem'    'Tennis'      [19]
  'Math'    'Baseball'    [ 5]
>>

[Fischer Zheng]

Sorry, I don't think this what I wanted. The answer should be [ 5 ? ? ? ? 5 ] Matrix A is the lookup table. V1 and V2 provides the inputs for the lookup.

You can assume all the values will be there so that there is not going to be a question mark.

answer [5 ...] I got from using V1(1) and V2(1) with the lookup table.

[dpb]

Then swap the arguments to ismember; will return the positions found in A for each row of the combined v...

If, as in this case, there's a bunch that aren't there the direct addressing won't work owing to return 0 but if they're all going to be there then it'll work without the loop to test for missing indices...see addendum above...

[Fischer Zheng]

With your approach, when lookup is unique, it is working fine. If lookup table is not unique, this will run into problems. For my purpose, this is good enough.

Thanks,

[dpb]

Well, yeah, that was part of the problem description. If it isn't, that's a fish of another kettle...

I'd note it'd probably be simpler also if you kept or made a hash key from the two variables as well as in the lookup table to avoid the need to mush the two together and using the 'rows' flag to get around the storage issues of cell arrays being needed for the disparate data types and the length issues in concatenating character data.