adjust pca metrics to use proportion, plot percent

Author: DWGodwin

gelos/analysis.py

@@ -262,11 +262,11 @@ def run_analysis(
                     )
 
                 cache_path = layer_dir / f"{prefix}_{met_type}.csv"
-                if cache_path.exists():
-                    logger.info(f"cached {met_type} result exists at {cache_path}, skipping")
-                else:
-                    met_fn = METRICS[met_type]
-                    met_fn(embeddings, output_dir=layer_dir, prefix=prefix, **met_params)
+                # if cache_path.exists():
+                #     logger.info(f"cached {met_type} result exists at {cache_path}, skipping")
+                # else:
+                met_fn = METRICS[met_type]
+                met_fn(embeddings, output_dir=layer_dir, prefix=prefix, **met_params)
 
             # --- Run plots ---
             for p_cfg in strategy_cfg.get("plots", []):

gelos/comp_plots.py

@@ -32,7 +32,7 @@ def pca_ablation_table(
     pivot = df.pivot_table(
         index="experiment",
         columns="threshold",
-        values="percent_of_total_components",
+        values="proportion_of_total_components",
         aggfunc="first",
     )
 
@@ -55,14 +55,14 @@ def pca_ablation_table(
         aggfunc="first",
     ).reindex(index=pivot.index, columns=pivot.columns)
 
-    # Annotate cells with percent and total count
+    # Annotate cells with proportion and total count
     for i in range(pivot.shape[0]):
         for j in range(pivot.shape[1]):
             pct = pivot.values[i, j]
             n = n_comp_pivot.values[i, j]
             if not np.isnan(pct):
                 ax.text(
-                    j, i, f"{int(pct)}%\n({int(n)})",
+                    j, i, f"{int(pct*100)}%\n({int(n)})",
                     ha="center", va="center", fontsize=9,
                 )
 

gelos/metrics.py

@@ -41,16 +41,14 @@ def pca_ablation(
             {
                 "threshold": threshold,
                 "n_components": pca.n_components_,
-                "percent_of_total_components": pca.n_components_ / D,
-                "proportion_of_total_components": f"{pca.n_components_} / {D}",
+                "proportion_of_total_components": pca.n_components_ / D,
                 "total_variance_explained": float(pca.explained_variance_ratio_.sum()),
             }
         )
         logger.info(
             f"PCA ablation: threshold={threshold}, "
             f"n_components={pca.n_components_}, "
-            f"percent_of_total_components={pca.n_components_ / D}, "
-            f"proportion_of_total_components={pca.n_components_} / {D}, "
+            f"proportion_of_total_components={pca.n_components_ / D}, "
             f"variance={pca.explained_variance_ratio_.sum():.4f}"
         )