[Tex/LaTex] pgfplots and linear regression with semilogyaxis

pgfplotspgfplotstable

I would like to add a linear regression line to my plot. The MWE is:

\begin{document}

\pgfplotstableread{
Year  ICumCDP
2008  3.73E-07
2009  3.35E-07
2010  2.01E-07
2011  3.88E-07
2012  1.36E-06
}{\tableICumCDP}
\pgfplotstablecreatecol[linear regression={ymode=log}, x=Year, y=ICumCDP]{regression}{\tableICumCDP}
\xdef\slope{\pgfplotstableregressiona}     % save the slope parameter
\xdef\intercept{\pgfplotstableregressionb} % save the intercept parameter

\begin{tikzpicture}
  \begin{semilogyaxis}[
    /pgf/number format/.cd,
    use comma,
    1000 sep={},
    log basis y = 10,
    ylabel = {ICumCDP},
    ymin = 1E-8,
    ymax = 1E-4,
    yminorgrids = true,
    xlabel = {Year},
    xtick = {2008,2009,2010,2011,2012}
]
    \addplot [only marks, color=blue, x=Year, y=ICumCDP] table {\tableICumCDP};
    \addplot [no markers, color=red, domain=2008:2012] {exp(x*\slope + \intercept)};
  \end{semilogyaxis}
\end{tikzpicture}\\
\end{document}

The printed regression line is not correct:

Picture

My code is an modification of How to expand linear regression fit to the full x-axis range while using semilogyaxis

I have asked the same question in a german forum, but nobody could help me till now.

Best Answer

This seems to happen because of numerical inaccuracies (large numbers for the year, small numbers for y). If you transform your data (which you can do on the fly), the regression works:

\documentclass[border=5mm]{standalone}
\usepackage{pgfplots, pgfplotstable}

\begin{document}

\pgfplotstableread{
Year  ICumCDP
2008  3.73E-07
2009  3.35E-07
2010  2.01E-07
2011  3.88E-07
2012  1.36E-06
}{\tableICumCDP}
\pgfplotstableset{
    create on use/shifted year/.style={ % Make the row number available
            create col/expr=\thisrow{Year}-2008
        }
}
\pgfplotstablecreatecol[linear regression={ymode=log, x=shifted year}]{regression}{\tableICumCDP}

\xdef\slope{\pgfplotstableregressiona}     % save the slope parameter
\pgfmathsetmacro\intercept{\pgfplotstableregressionb-\slope*2008}

\begin{tikzpicture}
  \begin{semilogyaxis}[
    /pgf/number format/.cd,
    use comma,
    1000 sep={},
    log basis y = 10,
    ylabel = {ICumCDP},
    ymin = 1E-8,
    ymax = 1E-4,
    yminorgrids = true,
    xlabel = {Year},
    xtick = {2008,2009,2010,2011,2012}
]
    \addplot [only marks, color=blue, x=Year, y=ICumCDP] table {\tableICumCDP};
    \addplot [no markers, color=red, domain=2008:2012] {exp((x)*\slope + \intercept)};
  \end{semilogyaxis}
\end{tikzpicture}\\
\end{document}
Related Question