Relative Error Minimization and result draw!

ROOT
1

Hello,

I am a new user, and i migrated from PAW but i need some help to understand some basic function of ROOT and if (and how) is it possible to do some function that i did with PAW

  1. I need to minimize the error not on an absolute value but on a relative value, in PAW i did this by dividing the output for 1/e_i^2 when e_1 is the difference between the real result and the calculated one. Is it possible to do something similar in ROOT?

  2. This is a stupid question i know, but i never really understood graph in PAW and it looks like i started with the same foot also in ROOT :smiley:
    I know how do graph distribution from the Input Tree, but what if i want to draw the results of the ANN, and for example i want to draw the error, or the average error given some set condition?

Thanks a lot for your help!
Max

2

Hello,

if I have understood well you point your first point you want to perform a least square (chi2) fit to a set of points with error. You can do this by using in ROOT the class TGraphErrors and then call the Fit method.

For the second point, would you like to know how to use get the error from the ANN class in ROOT or just to draw a graph from a set of values ?

Regards

Lorenzo

3

Not exactly, I want to train a neural net, but when training the net and trying to minimize the error, i don’t want the minimization done on the absolute error but on the relative error, because i have output that changes a lot, as a matter of fact in order for the net to process it, i had to apply a log(output) so that the values were on a better scale. Instead of that i’d like to keep the original values but minimize the relative error, not the absolute

No, i know how to draw graph, but i don’t know how to graph the error from the ANN and how to filter the data to be displayed too

4

Are you using the NN class of ROOT, TMultiLayerPerceptron ?

In this case, I think you have as training possibilities only what is available in the SetLearningMethod()

For the second question, you just need to get the errors from the NN. There is a method in the class, GetError() for retrieving the error value.

I hope this answer your question, if not I would need a more detailed description, a code example would be good, even in PAW.

Lorenzo