Make t-SNE always render (embedding fallback)

app.py

@@ -459,18 +459,28 @@ def plot_tsne(
 
     if representation == "LWM Embedding":
         embed_mask = sampled_df["tech_embedding"].apply(lambda x: x is not None)
-        if embed_mask.sum() < 5:
-            fig = go.Figure()
-            fig.update_layout(
-                title="No per-technology embeddings found for the selected filters.",
-                xaxis=dict(visible=False),
-                yaxis=dict(visible=False),
-            )
-            return fig
-        sampled_df = sampled_df.loc[embed_mask].reset_index(drop=True)
-        features = np.stack(sampled_df["tech_embedding"].values)
+        if embed_mask.sum() >= 5:
+            sampled_df = sampled_df.loc[embed_mask].reset_index(drop=True)
+            features = np.stack(sampled_df["tech_embedding"].values)
+            title_prefix = "t-SNE of LWM Embedding"
+        else:
+            # Fallback: use the full embedding vector so the UI doesn't go blank when
+            # per-expert metadata is missing in the artifact.
+            base_mask = sampled_df["embedding"].apply(lambda x: x is not None)
+            if base_mask.sum() < 5:
+                fig = go.Figure()
+                fig.update_layout(
+                    title="No embeddings available for the selected filters.",
+                    xaxis=dict(visible=False),
+                    yaxis=dict(visible=False),
+                )
+                return fig
+            sampled_df = sampled_df.loc[base_mask].reset_index(drop=True)
+            features = np.stack(sampled_df["embedding"].values)
+            title_prefix = "t-SNE of LWM Embedding (full vector)"
     else:
         features = build_tsne_raw_vectors(sampled_df["spectrogram"])
+        title_prefix = "t-SNE of Raw Spectrogram"
 
     if features.size == 0:
         fig = go.Figure()
@@ -510,7 +520,7 @@ def plot_tsne(
         y="y",
         color=color_column,
         hover_data=["tech", "snr", "mod", "mob"],
-        title=f"t-SNE of {representation} ({len(sampled_df)} samples)",
+        title=f"{title_prefix} ({len(sampled_df)} samples)",
         template="plotly_white",
         category_orders=category_orders,
     )