Merge pull request #418 from anyangml2nd/fix/barplot-datastructure

refactor: Reorient generalizability downstream score results to be ke…

Author: anyangml

lambench/metrics/results/pc_barplot.json

@@ -1,52 +1,38 @@
 {
-  "DPA-3.1-3M": {
-    "Inorganic Materials": 0.07,
-    "Molecules": 0.29,
-    "Catalysis": 0.53
-  },
-  "Orb-v3-omat": {
-    "Inorganic Materials": 0.07,
-    "Molecules": 0.42,
-    "Catalysis": 0.84
-  },
-  "SevenNet-MF-ompa": {
-    "Inorganic Materials": 0.06,
-    "Molecules": 0.53,
-    "Catalysis": 0.84
-  },
-  "MatterSim-v1-5M": {
-    "Inorganic Materials": 0.09,
-    "Molecules": 0.49,
-    "Catalysis": 0.91
-  },
-  "GRACE-2L-OAM": {
-    "Inorganic Materials": 0.06,
-    "Molecules": 0.48,
-    "Catalysis": 0.75
-  },
-  "MACE-MH-1": {
-    "Inorganic Materials": 0.07,
-    "Molecules": 0.22,
-    "Catalysis": 0.51
-  },
-  "DPA-3.2-5M": {
-    "Inorganic Materials": 0.06,
-    "Molecules": 0.17,
-    "Catalysis": 0.56
-  },
-  "DPA-2.4-7M": {
-    "Inorganic Materials": 0.13,
-    "Molecules": 0.4,
-    "Catalysis": 0.62
-  },
-  "MACE-MPA-0": {
-    "Inorganic Materials": 0.08,
-    "Molecules": 0.57,
-    "Catalysis": 0.76
-  },
-  "Orb-v2": {
-    "Inorganic Materials": 0.34,
-    "Molecules": 0.47,
-    "Catalysis": 0.85
+  "Inorganic Materials": {
+    "DPA-3.1-3M": 0.07,
+    "Orb-v3-omat": 0.07,
+    "SevenNet-MF-ompa": 0.06,
+    "MatterSim-v1-5M": 0.09,
+    "GRACE-2L-OAM": 0.06,
+    "MACE-MH-1": 0.07,
+    "DPA-3.2-5M": 0.06,
+    "DPA-2.4-7M": 0.13,
+    "MACE-MPA-0": 0.08,
+    "Orb-v2": 0.34
+  },
+  "Molecules": {
+    "DPA-3.1-3M": 0.29,
+    "Orb-v3-omat": 0.42,
+    "SevenNet-MF-ompa": 0.53,
+    "MatterSim-v1-5M": 0.49,
+    "GRACE-2L-OAM": 0.48,
+    "MACE-MH-1": 0.22,
+    "DPA-3.2-5M": 0.17,
+    "DPA-2.4-7M": 0.4,
+    "MACE-MPA-0": 0.57,
+    "Orb-v2": 0.47
+  },
+  "Catalysis": {
+    "DPA-3.1-3M": 0.53,
+    "Orb-v3-omat": 0.84,
+    "SevenNet-MF-ompa": 0.84,
+    "MatterSim-v1-5M": 0.91,
+    "GRACE-2L-OAM": 0.75,
+    "MACE-MH-1": 0.51,
+    "DPA-3.2-5M": 0.56,
+    "DPA-2.4-7M": 0.62,
+    "MACE-MPA-0": 0.76,
+    "Orb-v2": 0.85
   }
 }

lambench/metrics/vishelper/metrics_calculations.py

@@ -83,7 +83,7 @@ def calculate_generalizability_ood_error_metric(self) -> dict[str, float]:
             model: np.mean(values) for model, values in rearranged_m_bar_domain.items()
         }
 
-    def calculate_generalizability_downstream_score(self) -> dict[str, float]:
+    def calculate_generalizability_downstream_score(self) -> tuple[dict, dict]:
         raw_results = self.fetcher.fetch_downstream_results()
 
         # Extract necessary columns and prepare penalty dict
@@ -159,7 +159,7 @@ def calculate_generalizability_downstream_score(self) -> dict[str, float]:
         )  # drop models with missing domain results
 
         # # Now aggregate all domains to get the final generalizability metrics for each model
-        return domain_results, domain_results.mean(axis=1).to_dict()
+        return domain_results.to_dict(orient="index"), domain_results.mean(axis=1).to_dict()
 
     def calculate_stability_results(self) -> dict[str, float]:
         """This calculates the stability score for a given LAM."""