MATLAB: Error using varfun function. “Matrix dimensions must agree.”

dimensionsmatrixStatistics and Machine Learning Toolboxtablevarfun

Hello guys,
I am trying to do the following operation:
I want to multiply a table of 52824×9, named X, by the first column of a table of 9×4. The table's name is coef, the first column variable name is Estimate. Therefore I would like to do something like this: 
yPred=X*coef.Estimate
I am using the varfun function in the following way:
yPred=varfun(@(x) x.*coef.Estimate,X);
However, I am getting the following message:
Error using tabular/varfun>dfltErrHandler (line 433)
Applying the function '@(x)x.*coef.Estimate' to the variable 'VarName3' generated the following error:
Matrix dimensions must agree.
As I understand, I am multiplying 52824×9*9×1, so I don't know where my mistake is.
If anyone knows how to solve this, or could provide any solution it would be very much appreciated.
Many thanks,
Natalia

Best Answer

Several points:
  • "As I understand, I am multiplying 52824x9*9x1" No, you're using .*, the memberwise multiplication, not * the matrix multiplication. With .* the matrices must have the same size in all dimensions unless or one of them must be scalar in the mismatched dimensions.
  • "I want to multiply a table of 52824x9". Multiplication on tables are not defined. You could transform a table as long as all the variables are numeric matrices and multiply that but you cannot multiply tables. Typically tables contain heterogeneous data so multiplication wouldn't make sense.
  • varfun applies the given function to each variable of the table in turn. So the x that your function receives will be 52824x1. I don't think you want to use varfun.
  • The variables of your tables are day, month, period, holiday (which is just a 0/1 flag), etc. Does it even make sense to multiply variables like day or month, by your coeff variables?
  • It looks like you're trying to do a linear fit. Can you even do a linear fit when one of your variable is a 0/1 flag?
Related Question