Fix bug in eventstats and add test #587
Conversation
|
@aarchiba @luojing1211 These test failures are in either test_ell1h.py or test_fbx.py and seem to occur randomly. They have nothing to do with my changes. Somehow those tests are extremely fragile. Any idea what is going on? |
|
This is because the relative differences between the analytical derivative and numerical derivative are a little bit bigger than the old tolerance (0.00104 vs 0.001). I found this is interesting, py2.7 does not report the bug. |
|
So, do we increase the tolerance or fix the algorithm so they match more closely? |
|
I think we can do it now before we have a better way to test the derivative precision. |
|
@luojing1211 when you say "I think we can do it now", what is "it"? I'd like to not have those failing tests. |
|
This PR now includes fixes for using pulse numbers either from a par file or computed internally by PINT. I know I should have kept them separate.. sorry. |
|
As soon as issue #593 is fixed, I'll add a good unit test for pulse numbers, which I think was not there before. |
* nanograv/master: Blacken/isort Add unit test for models.Spindown.change_pepoch() Allow change_pepoch() to be used with toas=None Fix parfile writing Fix test_times fix erfautils Change numerical derivative step size for H3 Increase stepsize Fixed import bug for maskParameters Fixed import and path issue with examples
* nanograv/master: Blackening of previous PR
* nanograv/master: Replaced bad nuppi clock file with correct Nancay clock file
* nanograv/master: Work around np.longdouble bug. black new test file add mulitiple observatories test. Extract proper motion in an astropy 2 compatible way add more tests for TOAs building Fix astropy table time column lost location information issue Get change_posepoch() to pass test Add test for change_posepoch() Make model a fixture in test_change_pepoch() Fix broadcasting corner case Fix broadcasting of get_psr_coords() over epoch Add change_pepoch() Set ECL=IERS2010 in test par file so test passes Return velocities from get_psr_coords() Allow 'pm_lat' and 'pm_lon_coslat' keywords in PulsarEcliptic under astropy 2 Add test for ecliptic to ICRS conversion
|
OK, I'm going to merge this so that master will be passing all the tests. |
This fixes a bug in sig2sigma().
I added a test, but it is not great, and really we should add tests for many of the other routines in this file.