feat(trip-cluster): density-adaptive timestamp clustering + tag seeds + anchors

2026-06-27 17:16:07 +02:00
parent 351a001903
commit 76be1fb5af
2 changed files with 254 additions and 0 deletions
@@ -0,0 +1,62 @@
+from photoflow.core.models import Asset
+from app.clustering import cluster_assets
+
+
+def _a(i, taken, gps=False, city=None):
+    return Asset(immich_id=i, taken_at=taken,
+                 gps_lat=45.0 if gps else None, gps_lon=12.0 if gps else None,
+                 place_city=city)
+
+
+def test_seed_tag_forms_one_cluster_not_gap_split():
+    # Two assets months apart but sharing a trip tag -> ONE seeded cluster.
+    assets = [_a("a", "2019-06-01T10:00:00"), _a("b", "2019-09-01T10:00:00")]
+    tags = {"a": ["Italy 2019"], "b": ["Italy 2019"]}
+    clusters = cluster_assets(assets, tags, {"Italy 2019"})
+    assert len(clusters) == 1
+    c = clusters[0]
+    assert c.seed_tag == "Italy 2019" and sorted(c.member_ids) == ["a", "b"]
+    assert c.confidence >= 0.9 and c.suggested_name == "Italy 2019"
+
+
+def test_sparse_old_regime_splits_on_adaptive_threshold():
+    # ~1 day intra-trip gaps; trips separated by 10 days (< 14d hard cap),
+    # so only the adaptive rule can split them.
+    a = [_a(f"a{i}", f"2008-06-0{i+1}T12:00:00") for i in range(5)]   # Jun 1..5
+    b = [_a(f"b{i}", f"2008-06-1{i+5}T12:00:00") for i in range(3)]   # Jun 15..17
+    assets = a + b
+    tags = {x.immich_id: [] for x in assets}
+    clusters = cluster_assets(assets, tags, set())
+    assert len(clusters) == 2
+    assert sorted(clusters[0].member_ids) == ["a0", "a1", "a2", "a3", "a4"]
+
+
+def test_dense_recent_regime_splits_on_adaptive_threshold():
+    # Hourly bursts within a day; 2-day gap between days.
+    day1 = [_a(f"d{i}", f"2024-03-10T{10+i:02d}:00:00") for i in range(4)]
+    day3 = [_a(f"e{i}", f"2024-03-12T{10+i:02d}:00:00") for i in range(4)]
+    assets = day1 + day3
+    tags = {x.immich_id: [] for x in assets}
+    clusters = cluster_assets(assets, tags, set())
+    assert len(clusters) == 2
+    assert sorted(clusters[0].member_ids) == ["d0", "d1", "d2", "d3"]
+
+
+def test_location_anchor_names_and_gps_confidence():
+    assets = [_a("a", "2020-05-01T10:00:00", gps=True, city="Kiev"),
+              _a("b", "2020-05-01T12:00:00", gps=True, city="Kiev"),
+              _a("c", "2020-05-01T14:00:00", gps=True, city="Kiev"),
+              _a("d", "2020-05-01T16:00:00", gps=True, city="Kiev"),
+              _a("e", "2020-05-01T18:00:00", gps=True, city="Kiev")]
+    tags = {x.immich_id: [] for x in assets}
+    c = cluster_assets(assets, tags, set())[0]
+    assert c.suggested_name == "Kiev"
+    assert c.confidence > 0.6           # full GPS lifts confidence
+    assert c.kind_guess == "trip"
+
+
+def test_small_scattered_cluster_marked_everyday():
+    assets = [_a("a", "2015-01-01T10:00:00"), _a("b", "2015-01-01T11:00:00")]
+    tags = {"a": [], "b": []}
+    c = cluster_assets(assets, tags, set())[0]
+    assert c.kind_guess == "everyday"