Closed Kladdy closed 3 years ago
Related to #16
Christian: I think I wrote a generator for you at /mnt/md0/tmp
I only adapted the training method so far but I didn’t test it yet.
can you do that?
You should check that the noise level is the same as before and that the traces look similar to what you had before
The current implementation is also not great for testing because everything is randomized
you could just write another generator where you specify a file and event id and then get different random realization from this event only. This is what you would need to make these sky maps of the error contours that we talked about anyway, the code should show you how to add noise
you can use: https://github.com/nu-radio/radiotools/blob/master/radiotools/plthelpers.py#L458-L495
for getting it on such a nice skymap projection it is more complicated. You also need to have bins that are the same size in solid angle. The solution is HEALPix, and the corresponding python implementation: https://healpy.readthedocs.io/en/latest/
the relevant lines of code are shown here: https://stackoverflow.com/questions/50483279/make-a-2d-histogram-with-healpix-pixellization-using-healpy
for each direction, calculate the healpy index
indices = hp.ang2pix(nside, theta, phi)
then calculate the counts per pixel
idx, counts = np.unique(indices, return_counts=True)
and make the skymap
hpx_map = np.zeros(npix, dtype=int)
hpx_map[idx] = counts
which you can then plot via
hp.mollview(hpx_map)
probably takes you a while to read through the documentation to understand healpix but it is super useful. Let me know if you have any questions. As a start you can also just do a 2D histogram in theta and phi using the radiotools.
Take a noiseless data, add ~1000-10000 different noise realisations, ru through network and make skymap of the angle difference, how is it distrubuted?