Thanks for doing this. I'd be cautious, however, about applying models derived using this method without first thoroughly evaluating their predictive power. The method minimizes the sum of the squared residuals in a transformed space -- in this case a semilog space -- instead of a real one. As a paper on the subject states: "It is generally not a good idea to transform nonlinear systems into linear ones even when it is possible. Not only is it somewhat cumbersome, since you must handle each different function in a different way, but the error distribution and the statistical weight of the data points change after such transformations. If you don't deal with this properly, it can result in substantial inaccuracies." Marco S. Caceci and William P. Cacheris (Florida State University), "Fitting curves to data: the simplex algorithm is the answer." BYTE, May 1984, 340-362. I've seen cases where curve-fitting in transformed spaces led to badly flawed predictions (with a logistic function in one instance, a quadratic function in another).
Thanks for doing this. I'd be cautious, however, about applying models derived using this method without first thoroughly evaluating their predictive power. The method minimizes the sum of the squared residuals in a transformed space -- in this case a semilog space -- instead of a real one. As a paper on the subject states: "It is generally not a good idea to transform nonlinear systems into linear ones even when it is possible. Not only is it somewhat cumbersome, since you must handle each different function in a different way, but the error distribution and the statistical weight of the data points change after such transformations. If you don't deal with this properly, it can result in substantial inaccuracies." Marco S. Caceci and William P. Cacheris (Florida State University), "Fitting curves to data: the simplex algorithm is the answer." BYTE, May 1984, 340-362. I've seen cases where curve-fitting in transformed spaces led to badly flawed predictions (with a logistic function in one instance, a quadratic function in another).
Thanks, do you know how to evaluate the performance of Logarithmic Regression model ?
interesting, how is it the plot of the raw data when the relationship is log-log rather than lin-log? do you have a video for that? Thank you
ruclips.net/video/b6ohCdzOxiU/видео.html
Thank you so much ❤
Can we write the dependent variable with log? I tried but error occurred.
you can
Thank you. Very well explained.
thank you very much for these little lessons, they are very useful!!!