Question about TMLPAnalyzer

Hi,I have some question about the output of TMLPAnalyzer->DrawDInputs(), how to understand this graph? how to explain those curve? what does the abscissa and the y-axis mean?
Thanks!

DrawDInputs() draws something that approximates the distribution of the derivative of the NN w.r.t. each input. That quantity is recognized as one of the measures to determine key quantities in the network.

What is done is to vary one input around its nominal value and to see how the NN changes. This is done for each entry in the sample and produces a distribution.

What you can learn from that is:

  • is variable a really useful, or is my network insensitive to it ?
  • is there any risk of big systematic ? Is the network extremely sensitive to small variations of any of my inputs ?

As you might understand, this is to be considered with care and can serve as input for an “educated guess” when optimizing the network.

Cheers.

I have added Christophe’s comments in the doc: See
root.cern.ch/root/html/TMLPAnaly … rawDInputs

Rene

what does “w.r.t.” mean?

“with respect to”

x-axis is the derivative of the NN w.r.t. each input, y-axis is enevt number?

x-axis is the derivative of the NN w.r.t. each input, y-axis is the number of entries.

Thanks!