Added choleskyinv.jl unit
#5
Conversation
This unit implements a Gaussian elimination recursion to compute the Cholesky factorization and its inverse in one pass. For small matrices this is faster then using `cholesky` and inversing the Cholesky factor with LinearAlgebra.jl. The main function creates two Cholesky factorizations, inheriting all the methods from LinearAlgebra.
Codecov Report
@@ Coverage Diff @@
## master #5 +/- ##
==========================================
+ Coverage 83.63% 83.73% +0.10%
==========================================
Files 3 4 +1
Lines 385 412 +27
==========================================
+ Hits 322 345 +23
- Misses 63 67 +4
Continue to review full report at Codecov.
|
Added tests for `choleskyinv!` and `choleskyinv`
|
Hello, i made all the changes. How should i proceed now? |
|
The tests are failing. Can you fix them? |
|
On my computer the testing are all passing. The only thing didn't work was the |
|
I think i know where the problem comes from. I m trying to fix it. |
|
Hello, now it seems to work on Julia > 1.2. I am calling |
|
We can drop support for Julia v1.0 and 1.1. Modify Project.toml (and bump to v0.2) |
|
In Project.Toml it is v0.2.1 now. Do you mean 0.3 or 0.2.2? |
|
Has to be 0.3 since we are dropping support for some Julia versions |
|
Right, let me optimize the code a bit further before doing that. Also, what do you think to modify the function to check the dimansion of the matrices; if they are small go ahead with the |
Now it does not require any extra memory: - the ``L_1`` factor overwrites the strictly lower triangle of the input matrix - its complex-conjugate transpose inverse overwrites the stricly upper triangle of the input matrix - ``D`` overwrites the diagonal part of the input matrix
|
Hello, Travis build runs smoothly now. For codecov, i have never used it, so, i am not sure what to do with the failing check. |
In some situations the Cholesky factor and its inverse are both needed. This unit implements a Gaussian elimination recursion to compute the Cholesky factor and its inverse in one pass. On small matrices at the moment being this appears faster then using the standard Julia
choleskyfunction and inverting the result. The main function creates two Cholesky factorizations, thus they inherit all special methods from LinearAlgebra.I am having two problems at this time :
showmethod i would like to show the two factorizations calling the show method of LinearAlgebra, but i am currently not succeding.Also, if anybody knows more efficient way to initialize to the identity matrix the LowerTriangular and UpperTriangular matrices used by the algorithms, please let me know.