Upload folder using huggingface_hub

app.py

@@ -768,18 +768,30 @@ _METRIC_LABELS = {
 }
 
 
-def _build_grid_summary_figure(field=None, min_size=None):
-    """Heatmap of mean silhouette per (model, threshold).
+_OVERVIEW_METRICS = {
+    "silhouette_cosine": ("Silhouette Score",    True,  "RdYlGn",   "Mean<br>Silhouette"),
+    "davies_bouldin":    ("Davies-Bouldin Index", False, "RdYlGn_r", "Mean<br>Davies-Bouldin"),
+    "n_clusters":        ("N Clusters",           True,  "Blues",    "Mean<br>N Clusters"),
+}
+
+
+def _build_grid_summary_figure(field=None, min_size=None, metric="silhouette_cosine"):
+    """Heatmap of a clustering metric per (model, threshold).
     field=None/'__global__' → average over all fields.
-    min_size=None/'__all__' → average over all min community sizes.
+    min_size=None/'__all__'  → average over all min community sizes.
+    metric: one of silhouette_cosine | davies_bouldin | n_clusters.
     """
     if METRICS_DF.empty:
         return go.Figure()
+    if metric not in _OVERVIEW_METRICS:
+        metric = "silhouette_cosine"
+
+    metric_label, higher_better, colorscale, cb_title = _OVERVIEW_METRICS[metric]
 
     def _short(m):
         return m.split("__")[-1] if "__" in m else m
 
-    src = METRICS_DF.dropna(subset=["silhouette_cosine"])
+    src = METRICS_DF.dropna(subset=[metric])
     if field and field != "__global__":
         src = src[src["field"] == field]
         if src.empty:
@@ -794,7 +806,7 @@ def _build_grid_summary_figure(field=None, min_size=None):
 
     agg = (
         src
-        .groupby(["model", "threshold"])["silhouette_cosine"]
+        .groupby(["model", "threshold"])[metric]
         .mean()
         .reset_index()
     )
@@ -802,50 +814,54 @@ def _build_grid_summary_figure(field=None, min_size=None):
 
     pivot = agg.pivot_table(
         index="short_model", columns="threshold",
-        values="silhouette_cosine", aggfunc="mean",
+        values=metric, aggfunc="mean",
     )
-    # Sort rows: worst → best (best appears at top visually in Plotly heatmap)
-    row_order = pivot.max(axis=1).sort_values(ascending=True).index.tolist()
+    # Sort rows so the best model appears at the top
+    best_per_row = pivot.max(axis=1) if higher_better else pivot.min(axis=1)
+    row_order = best_per_row.sort_values(ascending=higher_better).index.tolist()
     pivot = pivot.loc[row_order]
 
-    # Embed star in the best cell's text (avoids categorical axis coordinate issues)
-    best_idx = agg["silhouette_cosine"].idxmax()
+    # Identify best cell and embed star in cell text
+    best_idx = agg[metric].idxmax() if higher_better else agg[metric].idxmin()
     best = agg.loc[best_idx]
     best_short = _short(best["model"])
     best_thresh = best["threshold"]
 
     rows_list = list(pivot.index)
-    cols_list = list(pivot.columns)
+    cols_list  = list(pivot.columns)
+    fmt = ".0f" if metric == "n_clusters" else ".3f"
     text = []
     for ri, row_name in enumerate(rows_list):
         row_text = []
         for ci, col_val in enumerate(cols_list):
             v = pivot.iloc[ri, ci]
             if pd.notna(v):
-                cell = f"★ {v:.3f}" if (row_name == best_short and col_val == best_thresh) else f"{v:.3f}"
+                cell = f"★ {v:{fmt}}" if (row_name == best_short and col_val == best_thresh) else f"{v:{fmt}}"
             else:
                 cell = ""
             row_text.append(cell)
         text.append(row_text)
 
+    field_label = "all fields"   if (not field    or field    == "__global__") else field.replace("_", " ").title()
+    min_label   = "all min-sizes" if (not min_size or min_size == "__all__")    else f"min={min_size}"
+    n_models    = len(pivot.index)
+
     fig = go.Figure(go.Heatmap(
         z=pivot.values,
         x=[str(c) for c in pivot.columns],
         y=list(pivot.index),
-        colorscale="RdYlGn",
+        colorscale=colorscale,
         text=text,
         texttemplate="%{text}",
         textfont=dict(size=11),
-        colorbar=dict(title=dict(text="Mean<br>Silhouette", font=dict(size=11)), thickness=14),
-        hovertemplate="Model: %{y}<br>Threshold: %{x}<br>Mean silhouette: %{z:.3f}<extra></extra>",
+        colorbar=dict(title=dict(text=cb_title, font=dict(size=11)), thickness=14),
+        hovertemplate=f"Model: %{{y}}<br>Threshold: %{{x}}<br>{metric_label}: %{{z:{fmt}}}<extra></extra>",
     ))
 
-    field_label = "all fields" if (not field or field == "__global__") else field.replace("_", " ").title()
-    min_label  = "all min-sizes" if (not min_size or min_size == "__all__") else f"min={min_size}"
-    n_models = len(pivot.index)
+    better_str = "higher = better" if higher_better else "lower = better"
     fig.update_layout(
         title=dict(
-            text=f"Grid Search Overview — Mean Silhouette · {field_label} · {min_label}  ★ = best",
+            text=f"Grid Search Overview — {metric_label} ({better_str}) · {field_label} · {min_label}  ★ = best",
             font=dict(size=13),
         ),
         xaxis=dict(title="Threshold", type="category", tickfont=dict(size=11)),
@@ -880,6 +896,17 @@ def grid_metrics_tab():
         dbc.Card([
             dbc.CardBody([
                 dbc.Row([
+                    dbc.Col([
+                        html.Label("Metric", className="fw-semibold small mb-1"),
+                        dcc.Dropdown(
+                            id="dd-summary-metric",
+                            options=[{"label": v[0], "value": k}
+                                     for k, v in _OVERVIEW_METRICS.items()],
+                            value="silhouette_cosine",
+                            clearable=False,
+                            style={"fontSize": "13px"},
+                        ),
+                    ], width=3),
                     dbc.Col([
                         html.Label("Field", className="fw-semibold small mb-1"),
                         dcc.Dropdown(
@@ -2646,11 +2673,12 @@ def download_cluster_table(n_clicks, rows):
 
 @app.callback(
     Output("metrics-summary-graph", "figure"),
+    Input("dd-summary-metric", "value"),
     Input("dd-summary-field", "value"),
     Input("dd-summary-minsize", "value"),
 )
-def update_summary_graph(field, min_size):
-    return _build_grid_summary_figure(field, min_size)
+def update_summary_graph(metric, field, min_size):
+    return _build_grid_summary_figure(field, min_size, metric)
 
 
 @app.callback(