search

xraypy / tdl

TDL
11 stars 6 forks source link

Tests does not work for tdl / modules / xrf / xrf_peaks.py #23

Open belegnar opened 11 years ago

belegnar commented 11 years ago 
$ python xrf_peaks.py 
Traceback (most recent call last):
  File "xrf_peaks.py", line 1051, in <module>
    test_fit()
  File "xrf_peaks.py", line 1009, in test_fit
    import _test_dat as test_dat
ImportError: No module named _test_dat
ttrainor commented 11 years ago

Try the following: xrf_peaks.py line 1009 should be changed to read:

from tdl.modules.xrf import _test_dat as test_dat

By the way, we still need to change this to use lsq routine from scipy, current mpfit is VERY slow. I believe Matt is working on this in larch, which has very good built in fitting capabilties....

newville commented 11 years ago

In fact, just today, I made some progress on XRF analysis with Larch. It now reads XRF spectra from MCA files much, much better, and can do a classic XRF background fit.

I've started looking at the xrf_model.py and xrf_peaks.py code, and these will be a little more work. Well, I have the emission line energies included already, so the lookup code is different.

I'd really like to to get rid of the tricks used for 'energy_flag', 'fwhm_flag' and even 'ampl_factor'. There should be a 'adjust_energy_scale=True'. For the FWHM trick, I'll have a flag to help set that the width should go as "A + B*sqrt(E)", but for the amplitudes, this will be much more robust if one just uses an algebraic constraint. In short, each XRF Peak will allow more fine-grained definition. And, the fitting will be faster and parameter bounds will be more robust....

It could take a while, but I'm trying to get a better XRF Viewer for our maps, so this is a good time to work through this.

ttrainor commented 11 years ago

Awesome, thanks Matt. Yeah, I agree getting rid of some of the flags will make it much more user friendly...

belegnar commented 11 years ago

i've found that line 1009 should be 

import _test_data as test_dat

but anyway there are a lot of other errors. it looks like function calls were refactored to dictionaries in params, but test code was not.