I'm doing this so I can extend the function a long ways out. The full table ends at 600 GeV in momentum and 552.25 GeV^2 in s. I need it to go out to 200TeV in momentum and maybe 400^2 GeV^2 in s. I need to go this far because the next step is stick it in an integral that doesn't like to converge (converges with
$1/P_{cutoff}^3$) and has a range of 8 orders of magnitude:
$$\int \frac{ds dP\cos(Pz)}{s+P^2}\sigma(s,P)$$
The stuff about Context doesn't make sense to me as I didn't ask about that. There's only one question here. I was trying to explain what it was you were looking at when you open the notebook. (I found the math typeface stuff, so I'm about to clarify the original post.)
When I say it suggests a thing, that means it came from the error message, specifically NetTrain::arrdiv. This suggestion is included in the error message for divergent gradients while training. Sometimes error messages include hints about how to fix them. The last time I caused it, I was trying to get the training to converge by giving it other measures (log(y), y', y"/y, and few others) of the data to try to fit, but the convergence problem was caused by excessive redundancy in the NN.
I did make a separate notebook to ensure that the idea was sound to start with and figure out how it should done (but I'll attach that here for your convenience). That's the other tread that I point to. It is much simpler. The alternative is to do this trick by hand for each momentum. That's not too bad past 30 GeV, but it isn't so good below that.
Attachments:
