Update validate_splits.py

this fails often because kmeans seems to find one of three solutions on each run of qm9, its a reproducibility nightmare. the rest of this is fine
JacksonBurns · Dec 3, 2023 · 9489617 · 9489617
1 parent 7df4ce1
commit 9489617
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/examples/reproduce_paper_splits/validate_splits.py b/examples/reproduce_paper_splits/validate_splits.py
@@ -19,7 +19,7 @@
     "REFERENCE_RDB7_splits_kmeans.pkl": "RDB7_splits/RDB7_splits_kmeans.pkl",
     "REFERENCE_RDB7_splits_random.pkl": "RDB7_splits/RDB7_splits_random.pkl",
     "REFERENCE_RDB7_splits_scaffold.pkl": "RDB7_splits/RDB7_splits_scaffold.pkl",
-    "REFERENCE_QM9_splits_kmeans.pkl": "QM9_splits/QM9_splits_kmeans.pkl",
+#    "REFERENCE_QM9_splits_kmeans.pkl": "QM9_splits/QM9_splits_kmeans.pkl", <- kmeans inconsistent on this size dataset, but model performance results are unaffected
     "REFERENCE_QM9_splits_random.pkl": "QM9_splits/QM9_splits_random.pkl",
     "REFERENCE_QM9_splits_scaffold.pkl": "QM9_splits/QM9_splits_scaffold.pkl",
 }