One solution that people sometimes use is a pseudo-log transform: x => sign(x) * log(1+abs(x)). One still might want to represent several different orders of magnitude on both positive and negative axes, or possibly deal with zeros when the differences between those values is not as important. Namely, values cannot be zero or negative, since the logarithm of those values are not defined for the real numbers. Log/Signed Log Scale with Linear ComponentĪ log-scaled axis does have its limitations, though. On the log scaled plot, however, you can tell what their values are (in exponential terms), and even see that the noise affects the smaller values more as a proportion of their value. For values of x until 9, you cannot tell what the difference between their y values on the linear-scaled plot. In the above example, I plotted some data that increases exponentially across x, with iid noise across the different points.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |