Count number of specific values in matrix

I have a large matrix, m, and am trying to count the number of a specific value (i.e. How many indexes are of the value 4?)

I tried using

val = sum(m == 4);

but I end up with val being a matrix/vector of numbers. I assume these numbers are from each column and should be added together for the total, so I tried another

num = sum(val == 4);

but then I just end up with another vector/matrix.

How can I do it?

[Dev]

What if the value is zero?, how do I count then?

[Walter Roberson]

@Dev

num = sum(val == 0, 'all');   %r2018b or later
num = sum(val(:) == 0);       %any version
num = nnz(~val);              %any version

Answer 1

[Walter Roberson]

sum(m(:) == 4)

[Romy Wolstencroft]

This worked perfectly for me. Thank you

[MathWorks Support Team]

An alternative syntax available in R2018b or later is sum(m==4,'all'). But for this simple problem colonizing the input with m(:) is likely to be faster.

[Andrew Frane]

Using the colon probably won't be faster to execute. If you use m(:) then Matlab has to do the additional step of creating a new array m(:) before doing the logical evaluation. In my tests, using 'all' is consistently faster than using the colon to vectorize the array.

Answer 2

[Kye Taylor]

Try this:

numberOfNonZeros = nnz(m==4);

Using nnz is more efficient than converting logicals to numeric, which is required to apply sum()

[Walter Roberson]

In the test I just did, the timings of sum() vs nnz() could not consistently tell the two cases apart. nnz() might possibly have been slightly faster, but the range of timings showed so much overlap that no real conclusion could be reached. It would make sense that nnz() could be faster, but I can't prove it at the moment. sum() on a large enough array could be dispatched to LAPACK after all.

[Andrew Frane]

I did a million iterations after defining m as a 1x1000000 vector, and nnz(m==4) beat sum(m==4) every time.

I did another million iterations after defining m as a 1000x1000 matrix, and nnz(m(:)==4) beat both sum(m(:)==4) and sum(m==4, 'all') every time.

So nnz does appear to be more efficient than sum.

Answer 3

[Sean de Wolski]

This could be done easily with histc() and unique() to get the number of each value:

uv = unique(x);
n  = histc(x,uv);

Or with unique() and accumarray():

[uv,~,idx] = unique(x);
n = accumarray(idx(:),1)

[Royi Avital]

Pay attention that histcount() won't have the same result as histc() above for this case (Difference at the end).

[Walter Roberson]

Royi is correct.

At the time the question was asked, histcounts did not exist.

The newer histcounts is recommended instead of histc()

In the where you pass in the bin edges, then histc() counts values that exactly match the upper limit separately, but histcounts counts them together with the previous bin.

Answer 4

[ntsh kr]

>> a

a =

     5     5     5     5     5     5     5     6     9    96
     5     3     9     5     2     7     5     6     2     1
     8     3     6     9     8     7     5     1     6     9

>> ans1=sum(a==5)

ans1 =

     2     1     1     2     1     1     3     0     0     0

>> b=sum(ans1)

b =

    11

[Manoj Payani]

Many Thanks - It works perfect.

Answer 5

[dipanka tanu sarmah]

along with this if you want to count the number of NaN ,(if there any) use nnz(isnan(m))

Answer 6

[vimal kumar chawda]

But if we want ot do for NaN and any numeric value in large matrix then ?

ans1=sum(a==5) so at this my value is numerical (which is not same all time) and other is NaN which is common. But i need to count only numerical value at particular value of x.,x2,x3...............x7000 which is on y axis.

-How many times y appear on the at particular value of x?

[Walter Roberson]

nnz(isnan(a))     %count nan

For a count of each value, see https://www.mathworks.com/matlabcentral/answers/37196-count-number-of-specific-values-in-matrix#answer_46452

Answer 7

[Patrick Benz]

How can I count the values in the second column of an array depending on the values in the column?

I've got an array that looks something like that:

400     0
396     0
392     1
400     0
396     1
400     1

and I want to know how often there is a "1" or a "0" next to a "400" or next to the other values.

I tried it with this way https://de.mathworks.com/matlabcentral/answers/37196-count-number-of-specific-values-in-matrix#answer_46452

but this only gives me the total numbers of "1" and "0" and how often there is a 392 in the first column.

[Image Analyst]

This doesn't look like an answer to @Corey Bullard so I think you should start your own question.

[Patrick Benz]

I did. under

https://de.mathworks.com/matlabcentral/answers/788869-klassieren-und-zaehlen-der-werte-eines-arrays-classifying-and-counting-the-values-of-an-array?s_tid=srchtitle

but hence this problem seemed simmilar to me I just wanted to extend my range

[Corey Bullard]

Another year, another answer to this decade old question.

[Walter Roberson]

Answered at https://www.mathworks.com/matlabcentral/answers/788869-klassieren-und-zahlen-der-werte-eines-arrays-classifying-and-counting-the-values-of-an-array#answer_665744