Fix tests after changes in maude #122
Conversation
|
Are there any downstream issues with these differences from the maude update? |
|
I don't know for sure, but I doubt it very much. The issue was that the values are not exactly the same, which is not uncommon with floating point types. The question is whether these tests need to be to full precision. |
|
Seems reasonable to me too, though I've forgotten what the default isclose tolerances are. |
| @@ -63,7 +63,8 @@ def test_vcdu_0_raw(): | |||
|
|
|||
| ref_data = test_data['686111007-686111017']['raw'] | |||
| data = maude_decom.get_raw_aca_packets(686111007, 686111017) | |||
| for key in ['TIME', 'MJF', 'MNF']: | |||
| assert np.all(np.isclose(data['TIME'], ref_data['TIME'])) | |||
There was a problem hiding this comment.
@jeanconn's point was a good one -- np.isclose uses a default relative tolerance of (IIRC) 1e-6, so that this ends up being a large number here. Best to set rtol=0, atol=1e-3 so we test the times are matching to a msec.
| @@ -203,7 +204,10 @@ def test_vcdu_vs_level0(): | |||
| table2 = maude_decom._get_aca_packets(raw, start, stop, | |||
| combine=True, adjust_time=True, calibrate=True) | |||
| for col, col2 in zip(table.itercols(), table2.itercols()): | |||
| assert np.all(col == col2) | |||
| if col.name in ['TIME', 'END_INTEG_TIME']: | |||
There was a problem hiding this comment.
I updated this to just do the 1e-3 atol on the time columns.
|
I forgot about this PR. I just checked that tests pass after @jeanconn and will merge+tag now. |
Description
It seems like in the latest maude update, some floating point values changed slightly. This caused chandra_aca tests to fail. This PR changes those tests so floating point values are checked using
np.isclose(a, b)instead ofa == b.Testing
Fixes #