GraphNeuralNetworks.jl
GraphNeuralNetworks.jl copied to clipboard
Fix test laplacian lambda max
With this PR I added the seed for rand_graph
.
Codecov Report
Merging #242 (55fd2b4) into master (1381a78) will increase coverage by
0.13%
. The diff coverage is93.75%
.
@@ Coverage Diff @@
## master #242 +/- ##
==========================================
+ Coverage 88.29% 88.42% +0.13%
==========================================
Files 16 16
Lines 1597 1624 +27
==========================================
+ Hits 1410 1436 +26
- Misses 187 188 +1
Impacted Files | Coverage Δ | |
---|---|---|
src/GNNGraphs/gnnheterograph.jl | 85.91% <93.75%> (+4.09%) |
:arrow_up: |
src/GNNGraphs/transform.jl | 96.55% <0.00%> (+0.43%) |
:arrow_up: |
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.
The problem this PR is trying to fix is the Laplacian test on nightly CI runs, e.g. here, right?
I don't know why that is happening but in principle computing the top eigenvalues of those rand bidirected graphs should not be problematic since their adjacency matrix is symmetric. Maybe zero-degree nodes cause some problems? Although I don't see why they should
Yes and also in the function _eigmax
there is the following line:
KrylovKit.eigsolve(Symmetric(A), x0, 1, :LR)[1][1]
hence A is forced to be symmetric. When I run the tests on my computer they run smoothly (I have Julia 1.8).
So all this PR does is remove a test...