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
Revise FIRuncFilter full covariance #190
Conversation
also addresses somewhat issue #175
- to check for full compatibility
- check coarse equivalence of covariances
add mean to random covariance matrix to lower condition number by some orders of magnitude
fixes issue with bad condition number of covariance
fixes issue of negative main diagonal in case="corr" (because resulting toeplitz matrix is not guaranteed to be positive semidefinite, which is bad for a covariance)
Codecov Report
@@ Coverage Diff @@
## master #190 +/- ##
==========================================
+ Coverage 44.60% 45.48% +0.87%
==========================================
Files 23 23
Lines 1641 1728 +87
Branches 285 313 +28
==========================================
+ Hits 732 786 +54
- Misses 834 852 +18
- Partials 75 90 +15
Continue to review full report at Codecov.
|
- enables faster calculation for fully certain theta with only diagonal information needed
- should ease future development
@BjoernLudwigPTB and I already had a in-depth look at the code structure, tweaked docstrings and discussed signatures of the new functions. As of now, the new
During our discussion, we were thinking of how to proceed from here. We would propose to let
@eichstaedtPTB : What do you think? |
My suggestion is to
|
also move old implementation to test
…RuncFilter_full_covariance
Alright, I made some final touches to this PR:
|
This pull request addresses issue #175 in some way.
By introducing a new function
_fir_filter
we can propagate full covariance information into the output of an FIR-filter. Based on this function a wrapperFIRuncFilter_2
is introduced, mimicing the behaviour of the existingFIRuncFilter
. (And thus preparing a potential replacement lateron.)Benefits of the new function(s):
Utheta == None
orUx is None
(this was already done for the FIRuncFilter, but in a much more complex way)A visual comparison with Monte Carlo covariance result show good agreement.
TODO: A quick runtime comparison of both methods should be done to evaluate potential performance gains/losses.