-
Notifications
You must be signed in to change notification settings - Fork 95
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[WIP, ENH] Add GODEC #120
[WIP, ENH] Add GODEC #120
Conversation
I looked for anything obvious, and it seems fine by me. I believe 'vn' is for variance normalization, rather than demeaning. I am not sure there I ever actually got the python implementation working correctly, though I do recall getting close. When I brought this up on the MEICA repository I was pointed towards the T1c outputs provided by MEICA, by @prantikk here. Limited testing suggested that the T1c mimicked the characteristics of the godec output from the 2018 Power et al paper, all with less fiddling. I believe the biggest problem I ran into was in the output from the algorithm, rather than running it. In any case, once this is tested it would be nice to compare the T1c data with the GoDec output to confirm (or not) similarities in a single dataset. |
MEDN time series was being constructed without the ignored components. Also the High-Kappa time series in T1c was being constructed without the mean.
Thanks for taking a look @dowdlelt. Also thanks for catching vn's meaning! Do you know if there's any kind of citation for the minimum image regression? The Power paper (as you know) recommended GODEC or a more standard GSR (which doesn't seem to be implemented in the same way in Certainly a quantitative comparison of the recommended method from the literature (GODEC) against the new, simpler method recommended here (T1c) would be nice for anyone who wants to publish using T1c. On a more general note, I think this is a good opportunity to move |
Codecov Report
@@ Coverage Diff @@
## master #120 +/- ##
=========================================
- Coverage 51.54% 49.9% -1.65%
=========================================
Files 32 33 +1
Lines 1969 2058 +89
=========================================
+ Hits 1015 1027 +12
- Misses 954 1031 +77
Continue to review full report at Codecov.
|
Still outputs to TED though.
# Conflicts: # tedana/workflows/t2smap.py
# Conflicts: # tedana/decomposition/eigendecomp.py # tedana/io.py # tedana/model/fit.py # tedana/selection/select_comps.py # tedana/tests/test_tedana.py # tedana/workflows/t2smap.py # tedana/workflows/tedana.py
# Conflicts: # tedana/decomposition/__init__.py # tedana/io.py # tedana/workflows/tedana.py
I remembered this PR while re-reading #79. With all of the merges there are no some conflicts. Would you mind updating this branch to be up-to-date with master? |
Closes #79. Ref #106.
@dowdlelt I know that you were able to get the GODEC implementation here working. I used that code as well, with several changes to remove redundancy and unused functions. Perhaps you could take a look at this PR?
Changes proposed in this pull request:
tedana.decomposition.godecomp.tedgodec
.ws_denoise
argument. Still haven't added a command-line argument.Fix potential bug in MEDN construction withinI've removed this to prevent bloat in the PR. I'm going to open a separate PR ([FIX] Include ignored components in ME-DN T1c time series #125) for this potential bug fix.tedana.utils.io.gscontrol_mmix
.ReplaceRemoved to reduce bloat.label
without_dir
and addout_dir
argument to many functions.To do: