Make cluster gcn reproducible #937
Conversation
![codeclimate[bot]](https://avatars.githubusercontent.com/in/9403?s=60&v=4){kind=small}
| """ | ||
|
|
||
| def __init__(self, G, clusters=1, q=1, lam=0.1, name=None): | ||
| def __init__(self, G, clusters=1, q=1, lam=0.1, name=None, seed=None): |
There was a problem hiding this comment.
Refactor this function to reduce its Cognitive Complexity from 17 to the 15 allowed.
|
Code Climate has analyzed commit d078921 and detected 0 issues on this pull request. View more on Code Climate. |
Codecov Report
@@ Coverage Diff @@
## develop #937 +/- ##
=========================================
+ Coverage 83.9% 84.3% +0.4%
=========================================
Files 53 53
Lines 5101 5103 +2
=========================================
+ Hits 4281 4301 +20
+ Misses 820 802 -18
Continue to review full report at Codecov.
|
Codecov Report
@@ Coverage Diff @@
## develop #937 +/- ##
=========================================
+ Coverage 84.3% 84.8% +0.5%
=========================================
Files 53 58 +5
Lines 5103 4958 -145
=========================================
- Hits 4301 4206 -95
+ Misses 802 752 -50
Continue to review full report at Codecov.
|
| targets = np.random.rand(petersen_graph.number_of_nodes(), target_size) | ||
| assert_reproducible( | ||
| lambda: cluster_gcn_nai(petersen_graph, targets, 4, 2, shuffle=shuffle) | ||
| ) |
There was a problem hiding this comment.
I feel that we need a test to also make sure that if the seed is different, then the models are indeed different. Otherwise, we have not established a causal link between the random seed value and initial model weights.
There was a problem hiding this comment.
Hi @kjun9
You might want to have a look at my suggestion for another test, otherwise this is good to merge.
P.
This updates cluster gcn's node generator to use the
random_stateutility that's now available, and adds some tests to make sure the model is reproducible when we set the global stellargraph seed.Part of #749