MATLAB: Comparing columns of cell arrays

cell arrayscomparefield

Hi,

I am new to matlab and would like to compare two cell arrays of a structure which have a different size.

My first field (drug.compound) of the structure (named drug) contains one column of IDs (28722×1) and my second field(drug.enzymes) contains one column of IDs 877×1. I would like to make a table in which the relation is shown between the IDs [28772×857] by a logical array.

Any suggestions?

I tried the strcmp function and == operator but not seem to work

Best Answer

As most of the data seem to be numeric encoded as character vectors, then we can easily convert them from char to numeric arrays and use an efficient logical comparison:

>> Nc = str2double(drugb.compounds(:));
>> Ne = str2double(drugb.enzymes(:)).';
>> X = bsxfun(@eq,Nc,Ne);
>> size(X)
ans =
   28772    1745

Some of the character vectors contain things like

"# Tanaka T, Tachibana H, Nonaka G, Nishioka I, Hsu FL, Kohda H, Tanaka O: Tannins and related compounds. CXXII. New dimeric, trimeric and tetrameric ellagitannins, lambertianins A-D, from Rubus lambertianus Seringe. Chem Pharm Bull (Tokyo). 1993 Jul;41(7):1214-20. ""Pubmed"":http://www.ncbi.nlm.nih.gov/pubmed/8374992 [Isolation]"

which will be converted to NaN and will therefore return false from the logical comparison.

Related Solutions

MATLAB: How do i compare cell array values and replace with string of code letters

Method one:

>> C = {'UUU','CUU','UUC','UUG'};
>> F = {'UUU','UUC'};
>> L = {'UUA','UUG','CUU','CUC','CUA','CUG'};
>> Xf = ismember(C,F)
Xf =
  1  0  1  0
>> Xl = ismember(C,L)
Xl =
  0  1  0  1

And then use indexing:

>> Y = Xf+2*Xl
Y =
   1   2   1   2
>> Z = 'FL';
>> Z(Y)
ans = FLFL

Or even something like:

>> char('F'*Xf + 'L'*Xl)
ans = FLFL

Use upper on ismember's input arguments to get case-insensitive comparisons.

Method two:

>> [~,X] = ismember(C,[F,L]);
>> V = cumsum(cellfun(@numel,{F,L}));
>> Y = 1+sum(V(:)<X)
Y =
   1   2   1   2

Method three:

>> C = {'UUU','CUU','UUC','UUG'};
>> F = {'UUU','UUC'};
>> L = {'UUA','UUG','CUU','CUC','CUA','CUG'};
>> F(2,:) = {'F'};
>> L(2,:) = {'L'};
>> D = [F,L];
>> [~,X] = ismember(C,D(1,:));
>> [D{2,X}]
ans = FLFL

MATLAB: Most efficient way to search in text arrays

Here are two possible approaches to accomplish this:

1) Use strcmp and find with string array

This option is O(n) for each lookup; in the worst case, every string in ids will be checked in order to find query. You can also use a cell array of character vectors for ids, in my test a string array was slightly faster.

>> ids = ["M1","M2", "M3"];
>> query = "M2";
>> index = find(strcmp(ids, query) == 1)
index = 
   2

2) Use containers.Map with char arrays as keys and indices as values

This option is O(n) to setup, but O(1) for each lookup. Meaning that regardless of how many ids are in your map, looking up each one will take the same amount of time.

>> ids = {'M1', 'M2', 'M3'};
>> indices = 1:length(ids);
>> idMap = containers.Map(ids, indices);
>> query = 'M2';
>> index = idMap(query)
index = 
    2

Here is a performance comparison. At each size increment the average time to compute 500 random queries was measured for each approach. Each approach used the same set of queries at each size increment. In my case, for less than ~1000 elements, find with strcmp is faster. But as the number of elements grows, containers.Map is the clear winner.

Best Answer

Related Solutions

MATLAB: How do i compare cell array values and replace with string of code letters

MATLAB: Most efficient way to search in text arrays

Related Question