search for standard sites
pub-search.waow.tech
search
zig
blog
atproto
fix: tune coarse clustering to avoid single-cluster collapse
min_cluster_size=100 with min_samples=5 collapsed to 2 clusters on
the current UMAP projection. Switch to min_cluster_size=50 with
min_samples=10 which produces ~64 coarse clusters with recognizable
topics.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
+2
-2
+2
-2
scripts/build-atlas
+2
-2
scripts/build-atlas
···
227
227
# --- step 3: clustering ---
228
228
import hdbscan
229
229
230
-
log("HDBSCAN coarse (min_cluster_size=100)...")
231
-
clusterer_coarse = hdbscan.HDBSCAN(min_cluster_size=100, min_samples=5)
230
+
log("HDBSCAN coarse (min_cluster_size=50)...")
231
+
clusterer_coarse = hdbscan.HDBSCAN(min_cluster_size=50, min_samples=10)
232
232
labels_coarse_raw = clusterer_coarse.fit_predict(X_2d)
233
233
n_coarse = len(set(labels_coarse_raw)) - (1 if -1 in labels_coarse_raw else 0)
234
234
n_outliers_coarse = (labels_coarse_raw == -1).sum()