Suppose you need to take a log transformation on set of data that is non-linear in nature.  You can tell this from a scatterplot.  Take ln (natural logarithm) of each x value and y value to complete the transformation. In doing so, you will get some errors in x for ln(0) and negative values for ln of values between 0 and 1. So to compensate for that add 1 to each x value before taking natural log. Then you will eliminate such problems and can get the regression equation and scatterplot accordingly