Update app.py

app.py

@@ -18,6 +18,7 @@ from distances_common import (
     sliced_wasserstein_distance_matrix,
 )
 from embed_raw import build_raw_embeddings
+    # noqa
 from embed_umap import build_umap_embeddings
 from embed_lwm import get_lwm_encoder, build_lwm_embeddings
 
@@ -30,8 +31,10 @@ def _log(msg: str):
     print(msg, flush=True)
 
 
-def _matrix_payload(np_mat: np.ndarray, labels: Optional[List[str]] = None) -> gr.Update:
-    """Return a safe gr.Dataframe update (headers match col_count)."""
+def _matrix_payload(np_mat: np.ndarray, labels: Optional[List[str]] = None):
+    """
+    Return a safe gr.update payload for a Dataframe (headers must match col_count).
+    """
     df = pd.DataFrame(np_mat)
     if labels is not None and len(labels) == df.shape[0]:
         df.index = labels
@@ -66,11 +69,12 @@ def _plot_heatmap(D: np.ndarray, labels: Optional[List[str]] = None) -> np.ndarr
     return img
 
 
-def _load_uploaded_files(file_objs: List[gr.File]) -> List[Tuple[torch.Tensor, Optional[torch.Tensor], str]]:
+def _load_uploaded_files(file_objs: List) -> List[Tuple[torch.Tensor, Optional[torch.Tensor], str]]:
     """Load uploaded datasets. Expect torch files with keys: 'channels', 'labels' (optional)."""
     out = []
     for f in (file_objs or []):
-        path = f.name if hasattr(f, "name") else f
+        # gr.Files can give strings (paths) or objects with .name
+        path = getattr(f, "name", f)
         try:
             obj = torch.load(path, map_location="cpu")
             ch = obj["channels"]
@@ -83,10 +87,11 @@ def _load_uploaded_files(file_objs: List[gr.File]) -> List[Tuple[torch.Tensor, O
 
 def _load_demo_files(paths: List[str]) -> List[Tuple[torch.Tensor, Optional[torch.Tensor], str]]:
     out = []
-    for p in paths:
+    for p in paths or []:
         try:
             ch, y = load_pt_dataset(p)
-            out.append((ch, y, os.path.basename(os.path.dirname(p))))  # scenario folder as name
+            # scenario folder name as label
+            out.append((ch, y, os.path.basename(os.path.dirname(p))))
         except Exception as e:
             _log(f"[WARN] Failed to load demo dataset {p}: {e}")
     return out
@@ -179,10 +184,11 @@ def refresh_demo_tasks():
 
 def refresh_demo_scenarios(task: str):
     if not task:
-        return gr.update(choices=[], value=None)
+        return gr.update(choices=[], value=[])
     files = list_demo_dataset_files(task)
-    # multi-select
-    return gr.update(choices=files, value=files[:3] if len(files) >= 3 else files)
+    # multi-select defaults
+    default = files[:3] if len(files) >= 3 else files
+    return gr.update(choices=files, value=default)
 
 
 def run_compute(
@@ -195,7 +201,7 @@ def run_compute(
     n_eval_per_dataset: int,
     demo_task: str,
     demo_files: List[str],
-    uploaded_files: List[gr.File],
+    uploaded_files: List,
     umap_mode: str,
     umap_n_components: int,
     umap_n_neighbors: int,
@@ -235,20 +241,20 @@ def run_compute(
 
     # UMAP cfg
     umap_kwargs = dict(
-        n_components=umap_n_components,
-        n_neighbors=umap_n_neighbors,
-        min_dist=umap_min_dist,
+        n_components=int(umap_n_components),
+        n_neighbors=int(umap_n_neighbors),
+        min_dist=float(umap_min_dist),
         metric=umap_metric,
-        spread=umap_spread,
-        learning_rate=umap_learning_rate,
+        spread=float(umap_spread),
+        learning_rate=float(umap_learning_rate),
         n_epochs=None if umap_n_epochs in [None, 0] else int(umap_n_epochs),
-        negative_sample_rate=umap_negative_sample_rate,
+        negative_sample_rate=int(umap_negative_sample_rate),
         init=umap_init,
-        densmap=umap_densmap,
-        set_op_mix_ratio=umap_set_op_mix_ratio,
-        local_connectivity=umap_local_connectivity,
-        repulsion_strength=umap_repulsion_strength,
-        random_state=umap_random_state,
+        densmap=bool(umap_densmap),
+        set_op_mix_ratio=float(umap_set_op_mix_ratio),
+        local_connectivity=float(umap_local_connectivity),
+        repulsion_strength=float(umap_repulsion_strength),
+        random_state=int(umap_random_state),
         target_metric="categorical",
         target_weight=0.5,
     )