Hugging Face's logo

GreenGenomicsLab
/
TARA-WorldModel-VICReg

Tabular Regression
PyTorch
self-supervised-learning
vicreg
joint-embedding
marine-biology
metagenomics
environmental-modeling
protein-domains
tara-oceans
pfam
Model card Files
xet
 Community
GreenGenomicsLab commited on
Commit
2949e8e
·
verified ·
1 Parent(s): 13c836d

Upload results/phase4_longhurst_comparison.txt with huggingface_hub

Browse files
Files changed (1) hide show
  1. results/phase4_longhurst_comparison.txt +202 -0
results/phase4_longhurst_comparison.txt ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Provenance:
2
+ # Script: /media/drn2/External/TARA-Oceans/03_analyses/WorldModelApp/scripts/t22_functional_biomes_longhurst_20260127_210000.py
3
+ # Input1: /media/drn2/External/TARA-Oceans/03_analyses/WorldModelApp/results/phase2_embeddings.npz
4
+ # Input2: /media/drn2/External/TARA-Oceans/03_analyses/WorldModelApp/data/consolidated_metadata.tsv
5
+ # Date: 2026-01-27 12:26:03
6
+ # Integrity Check: PASSED
7
+
8
+ FUNCTIONAL BIOME CLUSTERING & LONGHURST PROVINCE COMPARISON
9
+ ======================================================================
10
+
11
+ Dataset: 1,810 GPS-matched ocean samples with 32D z_env embeddings
12
+ Source: VICReg co-trained environment encoder (latent_dim=32, Phase 2 T14)
13
+
14
+ CLUSTERING METHOD SELECTION
15
+ ----------------------------------------
16
+ Selected: HDBSCAN
17
+ Config: {
18
+ "method": "HDBSCAN",
19
+ "min_cluster_size": 20,
20
+ "min_samples": 15,
21
+ "n_clusters": 23,
22
+ "n_noise": 500,
23
+ "noise_frac": 0.27624309392265195,
24
+ "silhouette": 0.5995908336938653
25
+ }
26
+ N clusters: 23
27
+ N noise: 500 (27.6%)
28
+ Silhouette score: 0.5996
29
+
30
+ HDBSCAN best: mcs=20, ms=15, k=23, noise=27.6%, sil=0.600
31
+ K-Means best: k=15, sil=0.382
32
+
33
+ K-MEANS SWEEP (ALL K VALUES)
34
+ ----------------------------------------
35
+ k silhouette inertia
36
+ 3 0.2535 35269.3
37
+ 4 0.2661 30940.4
38
+ 5 0.2938 27465.7
39
+ 6 0.2879 24708.9
40
+ 7 0.3104 22001.0
41
+ 8 0.3373 19998.5
42
+ 9 0.3375 18565.6
43
+ 10 0.3419 17370.6
44
+ 11 0.3358 16283.3
45
+ 12 0.3532 15272.4
46
+ 13 0.3514 14529.2
47
+ 14 0.3702 13564.0
48
+ 15 0.3820 12648.0
49
+
50
+ HDBSCAN SWEEP (VALID CONFIGS)
51
+ ----------------------------------------
52
+ mcs ms k noise% sil
53
+ 20 15 23 27.6% 0.600
54
+ 20 10 24 25.5% 0.592
55
+ 20 5 29 25.0% 0.588
56
+ 30 15 18 30.7% 0.571
57
+ 30 10 19 27.8% 0.555
58
+ 30 5 18 26.5% 0.524
59
+ 50 15 11 35.0% 0.445
60
+ 75 15 8 29.8% 0.365
61
+ 100 15 7 35.1% 0.357
62
+ 75 10 8 23.8% 0.326
63
+ 100 10 8 23.8% 0.326
64
+ 50 10 10 23.4% 0.315
65
+ 50 5 11 22.9% 0.312
66
+ 75 5 8 20.6% 0.304
67
+ 100 5 8 20.6% 0.304
68
+
69
+ AGREEMENT WITH BIOGEOGRAPHIC CLASSIFICATIONS
70
+ ----------------------------------------
71
+ Reference ARI NMI
72
+ Longhurst provinces 0.5031 0.7205
73
+ Longhurst biomes 0.2144 0.5002
74
+ Ocean basins 0.2898 0.6363
75
+
76
+ K=5 MATCHING LONGHURST BIOMES
77
+ ----------------------------------------
78
+ Silhouette: 0.294
79
+ ARI vs Longhurst biomes: 0.2467
80
+ ARI vs ocean basins: 0.1432
81
+ NMI vs Longhurst biomes: 0.2588
82
+
83
+ LONGHURST PROVINCE DISTRIBUTION
84
+ ----------------------------------------
85
+ NADR: 299 (16.5%)
86
+ MEDI: 242 (13.4%)
87
+ NPTG: 199 (11.0%)
88
+ PEQD: 197 (10.9%)
89
+ SPSG: 149 (8.2%)
90
+ NAST: 117 (6.5%)
91
+ ARAB: 110 (6.1%)
92
+ ISSG: 73 (4.0%)
93
+ SSTC: 68 (3.8%)
94
+ CALC: 59 (3.3%)
95
+ SATL: 53 (2.9%)
96
+ APLR: 50 (2.8%)
97
+ NPPF: 49 (2.7%)
98
+ BPLR: 38 (2.1%)
99
+ BENG: 29 (1.6%)
100
+ WTRA: 28 (1.5%)
101
+ REDS: 24 (1.3%)
102
+ INDE: 14 (0.8%)
103
+ CNRY: 12 (0.7%)
104
+
105
+ LONGHURST BIOME DISTRIBUTION
106
+ ----------------------------------------
107
+ Trades: 823 (45.5%)
108
+ Westerlies: 533 (29.4%)
109
+ Coastal_Enclosed: 266 (14.7%)
110
+ Coastal: 100 (5.5%)
111
+ Polar: 88 (4.9%)
112
+
113
+ CLUSTER CHARACTERIZATION
114
+ ----------------------------------------
115
+ Cluster n Dominant Basin Dominant Province Lat Range
116
+ 0 56 Arctic (50%) BPLR (50%) -77.8 to 78.6
117
+ 1 25 Atlantic (100%) SSTC (100%) -57.8 to -47.2
118
+ 2 96 Atlantic (100%) NADR (92%) -32.2 to 65.8
119
+ 3 41 Pacific (100%) NPTG (34%) -40.9 to 48.0
120
+ 4 102 Atlantic (100%) NADR (56%) -40.6 to 66.1
121
+ 5 115 Mediterranean (97%) MEDI (97%) -64.3 to 45.7
122
+ 6 21 Atlantic (100%) BENG (100%) -34.9 to -34.5
123
+ 7 107 Mediterranean (98%) MEDI (98%) 27.4 to 43.7
124
+ 8 42 Atlantic (100%) NADR (55%) 19.2 to 61.0
125
+ 9 59 Atlantic (100%) NADR (100%) 48.4 to 59.1
126
+ 10 34 Indian (85%) ARAB (47%) -26.1 to 29.5
127
+ 11 21 Red_Sea (52%) REDS (52%) -17.6 to 23.4
128
+ 12 53 Indian (100%) ARAB (100%) 14.4 to 20.8
129
+ 13 35 Indian (100%) ISSG (100%) -35.2 to -9.4
130
+ 14 26 Pacific (100%) NPTG (81%) -39.0 to 49.2
131
+ 15 39 Pacific (100%) CALC (54%) -39.6 to 35.4
132
+ 16 34 Pacific (100%) SPSG (100%) -29.7 to -21.2
133
+ 17 179 Pacific (100%) PEQD (84%) -16.0 to 25.6
134
+ 18 58 Pacific (100%) PEQD (50%) -31.9 to 35.0
135
+ 19 36 Atlantic (100%) SATL (53%) -21.1 to 4.1
136
+ 20 86 Atlantic (100%) NAST (79%) -30.2 to 37.0
137
+ 21 23 Atlantic (100%) NADR (61%) -44.6 to 44.1
138
+ 22 22 Atlantic (100%) SATL (59%) -36.0 to 39.2
139
+
140
+ CLUSTER x LONGHURST PROVINCE CROSS-TABULATION
141
+ ----------------------------------------
142
+ province APLR ARAB BENG BPLR CALC CNRY INDE ISSG MEDI NADR NAST NPPF NPTG PEQD REDS SATL SPSG SSTC WTRA
143
+ cluster
144
+ 0 28 0 0 28 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
145
+ 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 25 0
146
+ 2 0 0 5 0 0 0 0 0 0 88 0 0 0 0 0 0 0 0 3
147
+ 3 0 0 0 0 6 0 0 0 0 0 0 1 14 7 0 0 11 2 0
148
+ 4 0 0 2 0 0 4 0 0 0 57 20 0 12 0 0 5 0 1 1
149
+ 5 4 0 0 0 0 0 0 0 111 0 0 0 0 0 0 0 0 0 0
150
+ 6 0 0 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
151
+ 7 0 0 0 0 0 0 0 0 105 0 0 0 2 0 0 0 0 0 0
152
+ 8 0 0 0 0 0 0 0 0 0 23 2 0 17 0 0 0 0 0 0
153
+ 9 0 0 0 0 0 0 0 0 0 59 0 0 0 0 0 0 0 0 0
154
+ 10 0 16 0 0 0 0 3 10 0 0 0 0 0 0 5 0 0 0 0
155
+ 11 0 6 0 0 0 0 0 4 0 0 0 0 0 0 11 0 0 0 0
156
+ 12 0 53 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
157
+ 13 0 0 0 0 0 0 0 35 0 0 0 0 0 0 0 0 0 0 0
158
+ 14 0 0 0 0 0 0 0 0 0 0 0 3 21 0 0 0 2 0 0
159
+ 15 0 0 0 0 21 0 0 0 0 0 0 0 0 0 0 0 18 0 0
160
+ 16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 34 0 0
161
+ 17 0 0 0 0 5 0 0 0 0 0 0 0 22 151 0 0 1 0 0
162
+ 18 0 0 0 0 0 0 0 0 0 0 0 0 15 29 0 0 14 0 0
163
+ 19 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 0 0 17
164
+ 20 0 0 0 0 0 8 0 0 0 0 68 0 0 0 0 10 0 0 0
165
+ 21 0 0 0 0 0 0 0 0 0 14 0 0 0 0 0 0 0 9 0
166
+ 22 0 0 0 0 0 0 0 0 0 0 8 0 1 0 0 13 0 0 0
167
+
168
+ CLUSTER x OCEAN BASIN CROSS-TABULATION
169
+ ----------------------------------------
170
+ basin Arctic Atlantic Indian Mediterranean Pacific Red_Sea Southern
171
+ cluster
172
+ 0 28 0 0 0 0 0 28
173
+ 1 0 25 0 0 0 0 0
174
+ 2 0 96 0 0 0 0 0
175
+ 3 0 0 0 0 41 0 0
176
+ 4 0 102 0 0 0 0 0
177
+ 5 0 0 0 111 0 0 4
178
+ 6 0 21 0 0 0 0 0
179
+ 7 0 2 0 105 0 0 0
180
+ 8 0 42 0 0 0 0 0
181
+ 9 0 59 0 0 0 0 0
182
+ 10 0 0 29 0 0 5 0
183
+ 11 0 0 10 0 0 11 0
184
+ 12 0 0 53 0 0 0 0
185
+ 13 0 0 35 0 0 0 0
186
+ 14 0 0 0 0 26 0 0
187
+ 15 0 0 0 0 39 0 0
188
+ 16 0 0 0 0 34 0 0
189
+ 17 0 0 0 0 179 0 0
190
+ 18 0 0 0 0 58 0 0
191
+ 19 0 36 0 0 0 0 0
192
+ 20 0 86 0 0 0 0 0
193
+ 21 0 23 0 0 0 0 0
194
+ 22 0 22 0 0 0 0 0
195
+
196
+ INTERPRETATION
197
+ ----------------------------------------
198
+ 1. The HDBSCAN clustering identifies 23 functional biome clusters from the 32D z_env embedding.
199
+ 2. ARI=0.503 against Longhurst provinces indicates moderate agreement with classical biogeography.
200
+ 3. ARI=0.290 against ocean basins indicates weaker agreement with coarse geographic divisions.
201
+ 4. The VICReg-trained environment encoder captures biogeographically meaningful environmental structure.
202
+ 5. 500 noise points (27.6%) represent samples in transition zones between functional biomes.