Identifying and deleting columns from a 2D wave containing same value in all rows

Share icon

Hi, 

This probably has a pretty simple solution but I just couldn't get to it.. I have a 2D wave that contains some columns where all rows have the same value. For example, in my case, I need to delete all such columns from my 2D wave where all rows of the column read zero or nan. I'm looking for a simple way to do this. 

In a different situation, I had to nan all such rows where all columns had the same value in a row. I am attaching a code here that I put together to nan all such rows of a 2D wave where every column had the same value, but I'm pretty sure a better way exists to do this more elegantly with some library function. Kindly advise. 

Thanks! 

Function raw2Dwavecheck()  //This function sets all such rows to nan where the entire row has the same value.
wave cantons_all_pmf,cantons_all_err_pmf
variable i,j,n,temp,val
variable rowcount=dimsize(cantons_all_pmf,0)
variable colcount=dimsize(cantons_all_pmf,1)
for(i=0;i<rowcount;i+=1)
n=0
temp=0
val=cantons_all_pmf[i][0]    //taking the first value of the row and check the remaining columns against it
for(j=0;j<colcount;j+=1)
temp=cantons_all_pmf[i][j]/val    //checking the quotient for each column in a row
if(temp==1)
n=n+1
endif
endfor
print n/colcount
if(n/colcount>0.8)                //if more than 80% columns give a quotient of 1, nan the row
cantons_all_pmf[i][]=nan
cantons_all_err_pmf[i][]=nan
endif

endfor
end

 

I wrote this. It works but is not the most elegant since I am running two for loops. Wondering if there is a simpler non-loopy way to perform this task.  

Function delcols()   //function to delete all columns that contain only zeros
wave trialwave
variable i,j,colcount=dimsize(trialwave,1)
variable rowcount=dimsize(trialwave,0),n

for(j=colcount-1;j>=0;j-=1)
n=0
for(i=0;i<rowcount;i+=1)
if(trialwave[i][j]==0)
n=n+1
endif
endfor
print n,rowcount
if(n==rowcount)
deletepoints/M=1 j,1, trialwave
endif
endfor
end

 

October 25, 2021 at 08:46 am - Permalink

Hi Peeyush,

 

One technique it to use the finduplicates function.

FindDuplicates [flags ] srcWave
The FindDuplicates operation identifies duplicate values in a wave and optionally creates various output waves. srcWave  can be either numeric or text.
When srcWave  is numeric, the /DN, /INDX, /RN and /SN flags create output waves as described below. If you omit all of these flags then FindDuplicates does nothing.

Loop through columns, I would go through in reverse order, seeing if the /RN wave is one point and that is equal to zero.  If so delete that column and move to the next lower one.

Andy

October 25, 2021 at 09:50 am - Permalink

A simple way to detect if columns have constant value is, e.g., for the matrix 'ddd':

MatrixOP/O aa=varCols(ddd)

This should give you a 0 entry for each column that has a constant value.

The choice of what to do next depends on whether your data contain NaNs or not.  If not it is fairly easy to accomplish what you want in three lines of code...

You start by creating a scaling wave that has NaN for each 0 entry in the varCols wave:

matrixop/o aa=setNans(const(1,numcols(ddd),1),equal(greater(varcols(ddd),0),0))

Next you scale the input matrix:

So the code you need to execute is:

MatrixOP/O aa=zapNaNs(scaleCols(ddd,setNans(const(1,numcols(ddd),1),equal(greater(varcols(ddd),0),0))))
Variable np=DimSize(aa,0)
MatrixOP/O cleanOut=redimension(aa,numRows(ddd),int32(np/numRows(ddd)))

I hope this helps,

AG


 

October 25, 2021 at 10:03 am - Permalink