Merge branch 'claude/compassionate-varahamihira-c09c42'

Author: 28nitin07

ml/v2/kaggle_train.ipynb

@@ -100,39 +100,7 @@
    "id": "cell-3",
    "metadata": {},
    "outputs": [],
-   "source": [
-    "# ── Dataset pipeline ──────────────────────────────────────────────────────────\n",
-    "AUTOTUNE = tf.data.AUTOTUNE\n",
-    "\n",
-    "def augment(image, label):\n",
-    "    image = tf.image.random_flip_left_right(image)\n",
-    "    image = tf.image.random_brightness(image, 40.0)\n",
-    "    image = tf.image.random_contrast(image, 0.75, 1.25)\n",
-    "    image = tf.image.random_saturation(image, 0.75, 1.25)\n",
-    "    image = tf.image.random_crop(image, [int(INPUT_SIZE * 0.9), int(INPUT_SIZE * 0.9), 3])\n",
-    "    image = tf.image.resize(image, [INPUT_SIZE, INPUT_SIZE])\n",
-    "    image = tf.clip_by_value(image, 0, 255)\n",
-    "    return image, label\n",
-    "\n",
-    "def make_dataset(split, augment_fn=None, shuffle=False):\n",
-    "    ds = keras.utils.image_dataset_from_directory(\n",
-    "        os.path.join(DATASET_DIR, split),\n",
-    "        class_names=CLASSES,\n",
-    "        image_size=(INPUT_SIZE, INPUT_SIZE),\n",
-    "        batch_size=BATCH_SIZE,\n",
-    "        label_mode='categorical',\n",
-    "        shuffle=shuffle,\n",
-    "        seed=42,\n",
-    "    )\n",
-    "    if augment_fn:\n",
-    "        ds = ds.map(augment_fn, num_parallel_calls=AUTOTUNE)\n",
-    "    return ds.prefetch(AUTOTUNE)\n",
-    "\n",
-    "train_ds = make_dataset('train', augment_fn=augment, shuffle=True)\n",
-    "val_ds   = make_dataset('val')\n",
-    "test_ds  = make_dataset('test')\n",
-    "print('Datasets loaded. Class order:', CLASSES)"
-   ]
+   "source": "# ── Dataset pipeline ──────────────────────────────────────────────────────────\nAUTOTUNE = tf.data.AUTOTUNE\n\ndef augment(image, label):\n    image = tf.image.random_flip_left_right(image)\n    image = tf.image.random_brightness(image, 40.0)\n    image = tf.image.random_contrast(image, 0.75, 1.25)\n    image = tf.image.random_saturation(image, 0.75, 1.25)\n    image = tf.image.random_crop(image, [int(INPUT_SIZE * 0.9), int(INPUT_SIZE * 0.9), 3])\n    image = tf.image.resize(image, [INPUT_SIZE, INPUT_SIZE])\n    image = tf.clip_by_value(image, 0, 255)\n    return image, label\n\ndef make_dataset(split, augment_fn=None, shuffle=False):\n    ds = keras.utils.image_dataset_from_directory(\n        os.path.join(DATASET_DIR, split),\n        class_names=CLASSES,\n        image_size=(INPUT_SIZE, INPUT_SIZE),\n        batch_size=None,          # unbatched so augment receives (H, W, 3)\n        label_mode='categorical',\n        shuffle=shuffle,\n        seed=42,\n    )\n    if augment_fn:\n        ds = ds.map(augment_fn, num_parallel_calls=AUTOTUNE)\n    ds = ds.batch(BATCH_SIZE)\n    return ds.prefetch(AUTOTUNE)\n\ntrain_ds = make_dataset('train', augment_fn=augment, shuffle=True)\nval_ds   = make_dataset('val')\ntest_ds  = make_dataset('test')\nprint('Datasets loaded. Class order:', CLASSES)"
   },
   {
    "cell_type": "code",
@@ -391,4 +359,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 5
-}
+}