mcncm/dysart

dysart's Issues

spectrum_fit is really slow

Play around with a number of initial guess heuristics to see if you can get away with being a little simpler and faster, or if there are obvious optimizations for the included heuristic.

config bug: debug database created on `$ . config off`

noise tolerances from physical limits

One (relatively coarse, but probably provisionally acceptable) way to check that a fitting procedure has succeeded is that it has a satisfactorily low χ^2 statistic. Exactly what "satisfactorily low" means, however, depends on the data! An ideal fit should have a reduced χ^2 essentially equal (whp) to the rms noise on the signal. This is an easy comparison to make for randomly-generated signals, but not necessarily for sample data.
It would be nice to show that the reduced χ^2 of the fit approaches the output noise power expected of the actual signal chain characteristics. This would ensure that unit tests for fitting routines not only give some assurance that the code isn't broken, but also (and more importantly) that our measurements aren't deceiving us. Furthermore, it would add a level of redundancy to various routine sanity checks, and might help to catch "oops, I forgot about that filter stage"-type hardware bugs.

No error message if you try to save to db when server isn't running

Importing dummy_devices.py from top-level-directory doesn't include modules correctly

Path to the sample data is wrong when running fitting tests.

When running test_spectra.py, the sample data cannot be loaded. The reason is that the path to the sample data is wrong in dysart/tests/measurement/fitting/env.py. Instead of data_file_path = '../../sample_data/' it should be data_file_path = '../../../sample_data/'.

Add check-status feature to config