[R] Cannot rescale a constant/zero column error.

[Jason Fill]

Jim,

Thank you for your reply.  I guess I am just not understanding what the difference is.  If I run your example using the pcmatrix<-data.frame(Q1=sample(1:5,10,TRUE),Q2=sample(1:5,10,TRUE)) and print out pcmatrix I get:

	Q1	Q2
1	98	99

Then if I mirror my input file to be, would would be what the samples from above would have created, it would look like this:

ID,Q1,Q2
1,2,3

And I run the script printing out pcmatrix I get the exact same thing.

	Q1	Q2
1	98	99

So I would think they are basically the same representations, however one works when passed into the prcomp function and the other does not.  I am just trying to tweak a script that we had created for us (cannot find the guy now) so I can pass more columns of data.  I clearly really do not know what I am doing, but again thanks for you time and any direction you can point me in would be awesome.

Thanks

Jason

On Nov 1, 2012, at 3:57 AM, Jim Lemon wrote:

> On 10/31/2012 11:47 PM, fillay89 wrote:
>> I am trying to run the R Script below, I have actually simplified it to just
>> this part that is causing issues.  When I run this script I continue to get
>> an error that says "cannot rescale a constant/zero column to a unit
>> variance".  I cannot figure out what is going on here.  I have stripped down
>> my data file so it is more manageable so I can try to figure this out.
>> 
>> The data.txt file that is being read looks like this:
>> 
>> 
>> I have made this file very basic on purpose to see if I could get this to
>> work, but it is not working.  Of course once I get this to actually work I
>> will expand the data file to match the data I am actually using.
>> 
>> 
>> 
>> If I change the attribute in the prcomp function to scale=FALSE of course I
>> can run my script.  But if it is scaling...which is causing the issues, it
>> errors.
>> 
>> Any help would be GREATLY appreciated.
>> 
> Hi fillay89,
> You seem to be trying to run a principal component analysis on two numbers, which won't produce anything useful. If I make up some data that is more realistic:
> 
> pcmatrix<-data.frame(Q1=sample(1:5,10,TRUE),Q2=sample(1:5,10,TRUE))
> pcmatrix[,1] <- pcmatrix[,1] + 96
> pcmatrix[,2] <- pcmatrix[,2] + 96
> x.pca <- prcomp(pcmatrix,retx=TRUE,center=TRUE,scale=TRUE,cor=TRUE)
> x.pca
> Standard deviations:
> [1] 1.263267 0.635733
> 
> Rotation:
>         PC1        PC2
> Q1 0.7071068 -0.7071068
> Q2 0.7071068  0.7071068
> 
> it runs and I get a slightly more useful result. Running your code does bomb in exactly the way you describe, as does:
> 
> x.pca <- prcomp(data.frame(Q1=99,Q2=98), retx=TRUE, center=TRUE, scale=TRUE, cor=TRUE)
> Error in prcomp.default(data.frame(Q1 = 99, Q2 = 98), retx = TRUE, center = TRUE,  :
>  cannot rescale a constant/zero column to unit variance
> 
> Your problem is that you have a data frame with only one value (i.e. a constant) in each column.
> 
> Jim