Fix: handle dataset_all.npz bundle format (points as object array, targets7 as 2D)

CENG428_Neural_Networks_Homework_1_&_2.ipynb

@@ -104,71 +104,95 @@
         "train_dir = os.path.join(base_path, 'train')\n",
         "test_dir = os.path.join(base_path, 'test')\n",
         "\n",
-        "def load_npz_files(directory):\n",
-        "    \"\"\"Load all .npz files from a directory (robust to different key formats).\"\"\"\n",
-        "    files = sorted([f for f in os.listdir(directory) if f.endswith('.npz')])\n",
+        "def load_bundle_npz(filepath):\n",
+        "    \"\"\"Load a bundled .npz like dataset_all.npz (keys: points, targets7, metadata).\"\"\"\n",
+        "    data = np.load(filepath, allow_pickle=True)\n",
+        "    points_obj = data['points']        # (N,) object array of variable-length arrays\n",
+        "    targets = data['targets7']          # (N, 7)\n",
+        "    metadata_obj = data['metadata']     # (N,) object array of dicts\n",
+        "    \n",
+        "    points_list = list(points_obj)\n",
+        "    metadata_list = []\n",
+        "    for i in range(len(metadata_obj)):\n",
+        "        md_raw = metadata_obj[i]\n",
+        "        t = targets[i]\n",
+        "        md = {\n",
+        "            'exists': int(md_raw.get('exists', t[0])) if isinstance(md_raw, dict) else int(t[0]),\n",
+        "            'thickness_name': str(md_raw.get('thickness_name', {0:'thin',1:'medium',2:'thick'}.get(int(np.argmax(t[1:4])),'unknown'))) if isinstance(md_raw, dict) else {0:'thin',1:'medium',2:'thick'}.get(int(np.argmax(t[1:4])),'unknown'),\n",
+        "            'thickness_id': int(md_raw.get('thickness_id', np.argmax(t[1:4]))) if isinstance(md_raw, dict) else int(np.argmax(t[1:4])),\n",
+        "            'cx': float(md_raw.get('cx', t[4])) if isinstance(md_raw, dict) else float(t[4]),\n",
+        "            'cy': float(md_raw.get('cy', t[5])) if isinstance(md_raw, dict) else float(t[5]),\n",
+        "            'r':  float(md_raw.get('r',  t[6])) if isinstance(md_raw, dict) else float(t[6]),\n",
+        "            'n_points': int(md_raw.get('n_points', len(points_obj[i]))) if isinstance(md_raw, dict) else len(points_obj[i]),\n",
+        "        }\n",
+        "        metadata_list.append(md)\n",
+        "    return points_list, targets, metadata_list\n",
+        "\n",
+        "\n",
+        "def load_individual_npz(directory, files):\n",
+        "    \"\"\"Load individual sample .npz files (keys: points, target7, meta_*).\"\"\"\n",
         "    points_list = []\n",
         "    targets_list = []\n",
         "    metadata_list = []\n",
-        "    \n",
-        "    for i, f in enumerate(files):\n",
+        "    for f in files:\n",
         "        data = np.load(os.path.join(directory, f), allow_pickle=True)\n",
-        "        keys = list(data.keys())\n",
-        "        \n",
-        "        # Print keys of first file for debugging\n",
-        "        if i == 0:\n",
-        "            print(f'Keys in first file ({f}): {keys}')\n",
-        "        \n",
-        "        # Get points\n",
         "        points_list.append(data['points'])\n",
-        "        \n",
-        "        # Get target — try common key names\n",
-        "        if 'target7' in keys:\n",
-        "            targets_list.append(data['target7'])\n",
-        "        elif 'targets7' in keys:\n",
-        "            targets_list.append(data['targets7'])\n",
-        "        elif 'target' in keys:\n",
-        "            targets_list.append(data['target'])\n",
-        "        elif 'labels' in keys:\n",
-        "            targets_list.append(data['labels'])\n",
-        "        else:\n",
-        "            raise KeyError(f'Cannot find target key in {f}. Available keys: {keys}')\n",
-        "        \n",
-        "        # Build metadata dict from whatever keys are available\n",
-        "        md = {}\n",
-        "        target = targets_list[-1]\n",
-        "        # Existence: from meta or infer from target[0]\n",
-        "        if 'meta_exists' in keys:\n",
-        "            md['exists'] = int(data['meta_exists'])\n",
-        "        else:\n",
-        "            md['exists'] = int(target[0])\n",
-        "        # Thickness name\n",
-        "        if 'meta_thickness_name' in keys:\n",
-        "            md['thickness_name'] = str(data['meta_thickness_name'])\n",
-        "        else:\n",
-        "            tid = int(np.argmax(target[1:4]))\n",
-        "            md['thickness_name'] = {0: 'thin', 1: 'medium', 2: 'thick'}.get(tid, 'unknown')\n",
-        "        # Thickness id\n",
-        "        if 'meta_thickness_id' in keys:\n",
-        "            md['thickness_id'] = int(data['meta_thickness_id'])\n",
-        "        else:\n",
-        "            md['thickness_id'] = int(np.argmax(target[1:4]))\n",
-        "        # Circle params\n",
-        "        md['cx'] = float(data['meta_cx']) if 'meta_cx' in keys else float(target[4])\n",
-        "        md['cy'] = float(data['meta_cy']) if 'meta_cy' in keys else float(target[5])\n",
-        "        md['r']  = float(data['meta_r'])  if 'meta_r'  in keys else float(target[6])\n",
-        "        md['n_points'] = int(data['meta_n_points']) if 'meta_n_points' in keys else len(data['points'])\n",
+        "        targets_list.append(data['target7'])\n",
+        "        t = data['target7']\n",
+        "        keys = list(data.keys())\n",
+        "        md = {\n",
+        "            'exists': int(data['meta_exists']) if 'meta_exists' in keys else int(t[0]),\n",
+        "            'thickness_name': str(data['meta_thickness_name']) if 'meta_thickness_name' in keys else {0:'thin',1:'medium',2:'thick'}.get(int(np.argmax(t[1:4])),'unknown'),\n",
+        "            'thickness_id': int(data['meta_thickness_id']) if 'meta_thickness_id' in keys else int(np.argmax(t[1:4])),\n",
+        "            'cx': float(data['meta_cx']) if 'meta_cx' in keys else float(t[4]),\n",
+        "            'cy': float(data['meta_cy']) if 'meta_cy' in keys else float(t[5]),\n",
+        "            'r':  float(data['meta_r'])  if 'meta_r'  in keys else float(t[6]),\n",
+        "            'n_points': int(data['meta_n_points']) if 'meta_n_points' in keys else len(data['points']),\n",
+        "        }\n",
         "        metadata_list.append(md)\n",
-        "    \n",
         "    return points_list, np.array(targets_list), metadata_list\n",
         "\n",
+        "\n",
+        "def load_npz_files(directory):\n",
+        "    \"\"\"Auto-detect format and load .npz files from a directory.\"\"\"\n",
+        "    all_files = sorted([f for f in os.listdir(directory) if f.endswith('.npz')])\n",
+        "    print(f'Found {len(all_files)} .npz file(s) in {directory}')\n",
+        "    \n",
+        "    # Check if there's a bundle file (dataset_all.npz)\n",
+        "    bundle_files = [f for f in all_files if 'dataset' in f.lower() or 'all' in f.lower()]\n",
+        "    individual_files = [f for f in all_files if f.startswith('sample_')]\n",
+        "    \n",
+        "    if bundle_files:\n",
+        "        # Load from bundle (dataset_all.npz)\n",
+        "        bundle_path = os.path.join(directory, bundle_files[0])\n",
+        "        print(f'Loading bundle file: {bundle_files[0]}')\n",
+        "        return load_bundle_npz(bundle_path)\n",
+        "    elif individual_files:\n",
+        "        # Load individual sample files\n",
+        "        print(f'Loading {len(individual_files)} individual sample files')\n",
+        "        first_data = np.load(os.path.join(directory, individual_files[0]), allow_pickle=True)\n",
+        "        print(f'Keys in first file: {list(first_data.keys())}')\n",
+        "        return load_individual_npz(directory, individual_files)\n",
+        "    else:\n",
+        "        # Try all files as individual samples\n",
+        "        print(f'Loading {len(all_files)} files as individual samples')\n",
+        "        first_data = np.load(os.path.join(directory, all_files[0]), allow_pickle=True)\n",
+        "        keys = list(first_data.keys())\n",
+        "        print(f'Keys in first file: {keys}')\n",
+        "        # If it looks like a bundle\n",
+        "        if 'targets7' in keys and 'metadata' in keys:\n",
+        "            return load_bundle_npz(os.path.join(directory, all_files[0]))\n",
+        "        else:\n",
+        "            return load_individual_npz(directory, all_files)\n",
+        "\n",
+        "\n",
         "# Load training data\n",
         "all_train_points, all_train_targets, all_train_metadata = load_npz_files(train_dir)\n",
-        "print(f'Loaded {len(all_train_points)} training samples')\n",
+        "print(f'\\nLoaded {len(all_train_points)} training samples')\n",
         "print(f'Target shape: {all_train_targets.shape}')\n",
         "\n",
         "# Inspect a sample\n",
-        "print(f'\\nSample file: points shape={all_train_points[0].shape}')\n",
+        "print(f'\\nSample: points shape={all_train_points[0].shape}')\n",
         "print(f'Sample target: {all_train_targets[0]}')\n",
         "print(f'Sample metadata: {all_train_metadata[0]}')"
       ]