[dut_alignment] Binning in residuals vs. position hists

[DavidLP]

The binning shows very strong ´Moiré effects and should be optimized:

[laborleben]

There is the possibility to adjust the binning:

testbeam_analysis/testbeam_analysis/dut_alignment.py

Line 1080 in af3bb1b

# smaller devices needs None, otherwise npixels_per_bin=5 and nbins_per_pixel=1 might improve the first step

.
The coarse binning helps to improve the fit. The effect the plot is showing is by far not the worst.

The effect is probably both, a cobination of a binning effect (plotting) and a pixelized telescope and DUT.

[DavidLP]

If the fit result depends on the binnig then this is not correct and missuses the bin width as a frequency filter. Also I do not want to optimize bin width by hand for each device. Choosing the "best" bin size is a science itself (paper) and I was hoping scipy.binned_stat does that, but apparently only np.histogram has fancy algos for autobinning,

[laborleben]

We use "auto" binning if no binning value is given. The "auto" binning is part of numpy 1.11.

[DavidLP]

And auto uses Freedman Diaconis Estimator and sturges auto bin method , where I guess somebody thought about that.

[DavidLP]

Sorry Firefox missclicked