Error bands need a lot of throws
|Reported by:||cwilkinson||Owned by:||somebody|
|Priority:||major||Milestone:||Tuning paper, vxry|
I found that the error bands produced for even a single parameter are not very good for relatively small numbers of throws (100), for now this can be fixed by simply making more throws, although that is time intensive. In future, replacing the TProfile with a 1D hist for each bin, and fitting a Gaussian to the bin to get the error (for each y-axis bin) is probably better.
For the histograms used for fitting, a limited range (maybe +/-3 * RMS) and a variable number of bins (maybe sqrt(entries) would be wise.