cmccormick1
commented
4 years ago Testing/Debugging Process: Downloaded a copy of the 4/1/19 version (8668ef7) to test old results against the newer code.
Test 1. In the 4/1/19 version (8668ef7), put sorting of masses after finding valid indices (like how it is in the current version of library.py):
- File names in the library_debugging folder are called "TEST1_4_1_19_revised_massbin"
- No differences in bin indexing --> i.e. difference between the previous and current npz bin index arrays = 0 for all bins
Test 2. Reverted back to original sorting in 4/1/19 version (8668ef7) (i.e. sorted the list of masses before the valid_ind). Reran just for a sanity check; expecting them to be different.
- File names in the library_debugging folder are called "TEST2_4_1_19_revised_massbin"
- Got nonzero differences between the arrays.
- Traced through the code, I now think the problem is with the new version (061211b), not the old version (8668ef7).
- The ind_sort/ind array, which is the array that we get the bin indices from is relative to different things between different versions.
- In the 4/1/19 case (8668ef7), the indices are relative to temp_x, and in the 10/23/19 (061211b) case, the indices are relative to temp_x[valid_ind].
- Based on the tracing, line 101 (ind_sort = np.argsort(temp_x[valid_ind])) is wrong in the 10/23/19 (061211b) version because it gives the indices in the temp_x[valid_ind] array, not the indices of temp_x.
- Fixed it by replacing old (061211b) ind_sort with: argsort_valid_x = np.argsort(temp_x[valid_ind]) ind_sort = valid_ind[argsort_valid_x]
Test 3. Changed ind_sort in the current version (061211b) to be valid_ind[argsort_valid_x] where argsort_valid_x = np.argsort(temp_x[valid_ind]).
- File names in the library_debugging folder are called "TEST3_10_24_19_revised_massbin"
- Got mostly zeros when finding the difference between old and current versions. The ones that are not zero are either +1 or -1. *Some of the mass values are switched with the one next to it because they have the same mass values, giving the +1 or -1 difference. Even with this difference, I was able to reproduce the same results as before.
Versions after 4/1/19 through 10/23/19 have incorrect relative indexing (starting in 8668ef7). The indices in those versions are the indices of the elements in the valid index array, not the actual valid index elements themselves. Versions before 4/1/19 and after 10/23/19 (fixed in 10/24/19 version: 979c84f) have indices of sources relative to passed in data table. The change that fixed it is in lines 101-103 in 979c84f.