Closed JakobAsslaender closed 2 years ago
Merging #52 (52aa0f0) into master (2ba16fb) will increase coverage by
0.11%
. The diff coverage is87.50%
.
@@ Coverage Diff @@
## master #52 +/- ##
==========================================
+ Coverage 64.50% 64.61% +0.11%
==========================================
Files 63 63
Lines 3048 3069 +21
==========================================
+ Hits 1966 1983 +17
- Misses 1082 1086 +4
Impacted Files | Coverage Δ | |
---|---|---|
src/Tools/CoilSensitivity.jl | 67.51% <87.50%> (+2.07%) |
:arrow_up: |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update 2ba16fb...52aa0f0. Read the comment docs.
Hi, I switched the voxel-wise SVD to an Eigen-decomposition. This has speed and memory benefits, and maybe also fixes #46 . The memory benefit arises from never storing an
Nx x Ny x Nz x Nc x Nvectors
array, whereNvectors
is the number of singular vectors that span the null-space. Instead, I compute an array of sizeNx x Ny x Nz x Nc x Nc
directly, which is usually much smaller. On large datasets this makes the difference if it fits in the memory or not...I added the option to perform the Eigen-decomposition via power iterations, if only one set of maps is requested, and I set this flag to true by default.
On the test data sets and my laptop, the algorithm is about 6x faster compared to he old code.