class FeatureGen(tf.Module):\n def __init__(self):\n super(FeatureGen, self).__init__()\n self.meta_data = metaData\n self.landMark = [0, 468, 489, 522, 543]\n self.INPUT_SIZE = 32\n\n def fill_nans(self,X, l, r):\n z = X[:, l : r]\n nan_idx, _ = tf.unique(tf.where(tf.reduce_all(tf.math.is_nan(z), axis=1))[:, 0])\n not_nan_idx = tf.sets.difference(tf.reshape(tf.range(0, z.shape[0], dtype = tf.dtypes.int64), (1, -1)), tf.reshape(nan_idx, (1, -1))).values\n if (nan_idx.shape[0]==None):\n return z\n if (nan_idx.shape[0] * 4 > not_nan_idx.shape[0]):\n indices = tf.where(tf.math.is_nan(z))\n z = tf.tensor_scatter_nd_update(\n z,\n indices,\n tf.zeros((tf.shape(indices)[0]))\n )\n return z\n if nan_idx[0] == 0:\n k,l = not_nan_idx[0],not_nan_idx[1]\n if (k == l):\n print(f\"Uncaught exception, k can't be equal to l : k = {k}, l = {l}\")\n exit()\n d = (z[l] - z[k]) / (l-k)\n a = z[k]\n z[0] = a - d*k\n not_nan_idx=tf.concat([[0], not_nan_idx], axis = 0)\n nan_idx=nan_idx[1:]\n if (nan_idx.shape[0] == 0):\n return z\n if nan_idx[-1] == z.shape[0] -1 :\n k,l = not_nan_idx[-1],not_nan_idx[-2]\n if (k == l):\n print(f\"Uncaught exception, k can't be equal to l : k = {k}, l = {l}\")\n exit()\n d = (z[k] - z[l]) / (k-l)\n a = z[k]\n z[-1] = a + d*(z.shape[0] -1 - k)\n not_nan_idx=tf.concat([not_nan_idx,[z.shape[0]-1]], axis = 0)\n nan_idx = nan_idx[:-1]\n for i in nan_idx:\n k,l = not_nan_idx[not_nan_idx<i][-1],not_nan_idx[not_nan_idx>i][0]\n if (k == l):\n print(f\"Uncaught exception, k can't be equal to l : k = {k}, l = {l}\")\n exit()\n d = (z[l] - z[k]) / (l-k)\n a = z[k]\n z[i] = a + d*(i-k)\n return z\n\n def padding(self,data):\n N_FRAMES = data.shape[0]\n N_COLS = data.shape[1] # Number of Landmark indices i.e. 543\n N_DIMS = data.shape[2]\n\n if N_FRAMES < self.INPUT_SIZE:\n # Pad Data With Zeros\n pad_size = self.INPUT_SIZE - N_FRAMES\n pad_left = (pad_size // 2)\n pad_right = (pad_size // 2) + pad_size % 2\n\n # Pad By Concatenating Left/Right Edge Values\n data = tf.pad(data, tf.constant([[pad_left, pad_right], [0, 0], [0, 0]]), 'constant', constant_values=0)\n else:\n if N_FRAMES < self.INPUT_SIZE**2:\n # Repeat\n repeats = (self.INPUT_SIZE * self.INPUT_SIZE) // N_FRAMES\n data = tf.repeat(data, repeats=repeats, axis=0)\n # Pad To Multiple Of Input Size\n pool_size = len(data) // self.INPUT_SIZE + bool(len(data) % self.INPUT_SIZE)\n pad_size = (pool_size * self.INPUT_SIZE) % len(data)\n\n # Pad Start/End with Start/End value\n pad_left = (pad_size // 2)\n pad_right = (pad_size // 2) + pad_size % 2\n data = tf.pad(data, tf.constant([[pad_left, pad_right], [0, 0], [0, 0]]), 'constant', constant_values=0)\n # Reshape to Mean Pool\n data = tf.reshape(data, (self.INPUT_SIZE, -1, N_COLS, N_DIMS))\n # Mean Pool\n data = tf.experimental.numpy.nanmean(data, axis=1)\n return data\n\n @tf.function(input_signature=[\n tf.TensorSpec(shape=[None, 543, 3], dtype=tf.float32, name='inputs')\n ])\n def call(self, x):\n # shape -> None , 543 , 3\n for i in range(4):\n y = self.fill_nans(x,self.landMark[i],self.landMark[i+1])\n if i ==0:\n output = y\n continue\n output = tf.concat([output,y],axis=1)\n x = self.padding(output)\n return x\nTypeError Traceback (most recent call last)\n/tmp/ipykernel_27/1088593474.py in <module>\n----> 1 model.call(x).shape\n\n/opt/conda/lib/python3.7/site-packages/tensorflow/python/util/traceback_utils.py in error_handler(*args, **kwargs)\n 151 except Exception as e:\n 152 filtered_tb = _process_traceback_frames(e.__traceback__)\n--> 153 raise e.with_traceback(filtered_tb) from None\n 154 finally:\n 155 del filtered_tb\n\n/tmp/__autograph_generated_filecbah15ng.py in tf__call(self, x)\n 59 output = ag__.Undefined('output')\n 60 ag__.for_stmt(ag__.converted_call(ag__.ld(range), (4,), None, fscope), None, loop_body, get_state_2, set_state_2, ('output',), {'iterate_names': 'i'})\n---> 61 x = ag__.converted_call(ag__.ld(self).padding, (ag__.ld(output),), None, fscope)\n 62 try:\n 63 do_return = True\n\n/tmp/__autograph_generated_filev3fkbxo7.py in tf__padding(self, data)\n 58 pad_right = ag__.Undefined('pad_right')\n 59 repeats = ag__.Undefined('repeats')\n---> 60 ag__.if_stmt((ag__.ld(N_FRAMES) < ag__.ld(self).INPUT_SIZE), if_body_1, else_body_1, get_state_1, set_state_1, ('data',), 1)\n 61 try:\n 62 do_return = True\n\nTypeError: in user code:\n\n File \"/tmp/ipykernel_27/600884451.py\", line 98, in call *\n x = self.padding(output)\n File \"/tmp/ipykernel_27/3282046787.py\", line 60, in padding *\n if N_FRAMES < self.INPUT_SIZE:\n\n TypeError: '<' not supported between instances of 'NoneType' and 'int'\nclass FeatureGen(tf.Module):\n def __init__(self):\n super(FeatureGen, self).__init__()\n self.meta_data = metaData\n self.landMark = [0, 468, 489, 522, 543]\n self.INPUT_SIZE = 32\n\n def fill_nans(self,X, l, r):\n z = X[:, l : r]\n nan_idx, _ = tf.unique(tf.where(tf.reduce_all(tf.math.is_nan(z), axis=1))[:, 0])\n not_nan_idx = tf.sets.difference(tf.reshape(tf.range(0, z.shape[0], dtype = tf.dtypes.int64), (1, -1)), tf.reshape(nan_idx, (1, -1))).values\n if (nan_idx.shape[0]==None):\n return z\n if (nan_idx.shape[0] * 4 > not_nan_idx.shape[0]):\n indices = tf.where(tf.math.is_nan(z))\n z = tf.tensor_scatter_nd_update(\n z,\n indices,\n tf.zeros((tf.shape(indices)[0]))\n )\n return z\n if nan_idx[0] == 0:\n k,l = not_nan_idx[0],not_nan_idx[1]\n if (k == l):\n print(f\"Uncaught exception, k can't be equal to l : k = {k}, l = {l}\")\n exit()\n d = (z[l] - z[k]) / (l-k)\n a = z[k]\n z[0] = a - d*k\n not_nan_idx=tf.concat([[0], not_nan_idx], axis = 0)\n nan_idx=nan_idx[1:]\n if (nan_idx.shape[0] == 0):\n return z\n if nan_idx[-1] == z.shape[0] -1 :\n k,l = not_nan_idx[-1],not_nan_idx[-2]\n if (k == l):\n print(f\"Uncaught exception, k can't be equal to l : k = {k}, l = {l}\")\n exit()\n d = (z[k] - z[l]) / (k-l)\n a = z[k]\n z[-1] = a + d*(z.shape[0] -1 - k)\n not_nan_idx=tf.concat([not_nan_idx,[z.shape[0]-1]], axis = 0)\n nan_idx = nan_idx[:-1]\n for i in nan_idx:\n k,l = not_nan_idx[not_nan_idx<i][-1],not_nan_idx[not_nan_idx>i][0]\n if (k == l):\n print(f\"Uncaught exception, k can't be equal to l : k = {k}, l = {l}\")\n exit()\n d = (z[l] - z[k]) / (l-k)\n a = z[k]\n z[i] = a + d*(i-k)\n return z\n\n def padding(self,data):\n N_FRAMES = data.shape[0]\n N_COLS = data.shape[1] # Number of Landmark indices i.e. 543\n N_DIMS = data.shape[2]\n\n if N_FRAMES < self.INPUT_SIZE:\n # Pad Data With Zeros\n pad_size = self.INPUT_SIZE - N_FRAMES\n pad_left = (pad_size // 2)\n pad_right = (pad_size // 2) + pad_size % 2\n\n # Pad By Concatenating Left/Right Edge Values\n data = tf.pad(data, tf.constant([[pad_left, pad_right], [0, 0], [0, 0]]), 'constant', constant_values=0)\n else:\n if N_FRAMES < self.INPUT_SIZE**2:\n # Repeat\n repeats = (self.INPUT_SIZE * self.INPUT_SIZE) // N_FRAMES\n data = tf.repeat(data, repeats=repeats, axis=0)\n # Pad To Multiple Of Input Size\n pool_size = len(data) // self.INPUT_SIZE + bool(len(data) % self.INPUT_SIZE)\n pad_size = (pool_size * self.INPUT_SIZE) % len(data)\n\n # Pad Start/End with Start/End value\n pad_left = (pad_size // 2)\n pad_right = (pad_size // 2) + pad_size % 2\n data = tf.pad(data, tf.constant([[pad_left, pad_right], [0, 0], [0, 0]]), 'constant', constant_values=0)\n # Reshape to Mean Pool\n data = tf.reshape(data, (self.INPUT_SIZE, -1, N_COLS, N_DIMS))\n # Mean Pool\n data = tf.experimental.numpy.nanmean(data, axis=1)\n return data\n\n @tf.function(input_signature=[\n tf.TensorSpec(shape=[None, 543, 3], dtype=tf.float32, name='inputs')\n ])\n def call(self, x):\n # shape -> None , 543 , 3\n for i in range(4):\n y = self.fill_nans(x,self.landMark[i],self.landMark[i+1])\n if i ==0:\n output = y\n continue\n output = tf.concat([output,y],axis=1)\n x = self.padding(output)\n return x\nTypeError Traceback (most recent call last)\n/tmp/ipykernel_27/1088593474.py in <module>\n----> 1 model.call(x).shape\n\n/opt/conda/lib/python3.7/site-packages/tensorflow/python/util/traceback_utils.py in error_handler(*args, **kwargs)\n 151 except Exception as e:\n 152 filtered_tb = _process_traceback_frames(e.__traceback__)\n--> 153 raise e.with_traceback(filtered_tb) from None\n 154 finally:\n 155 del filtered_tb\n\n/tmp/__autograph_generated_filecbah15ng.py in tf__call(self, x)\n 59 output = ag__.Undefined('output')\n 60 ag__.for_stmt(ag__.converted_call(ag__.ld(range), (4,), None, fscope), None, loop_body, get_state_2, set_state_2, ('output',), {'iterate_names': 'i'})\n---> 61 x = ag__.converted_call(ag__.ld(self).padding, (ag__.ld(output),), None, fscope)\n 62 try:\n 63 do_return = True\n\n/tmp/__autograph_generated_filev3fkbxo7.py in tf__padding(self, data)\n 58 pad_right = ag__.Undefined('pad_right')\n 59 repeats = ag__.Undefined('repeats')\n---> 60 ag__.if_stmt((ag__.ld(N_FRAMES) < ag__.ld(self).INPUT_SIZE), if_body_1, else_body_1, get_state_1, set_state_1, ('data',), 1)\n 61 try:\n 62 do_return = True\n\nTypeError: in user code:\n\n File \"/tmp/ipykernel_27/600884451.py\", line 98, in call *\n x = self.padding(output)\n File \"/tmp/ipykernel_27/3282046787.py\", line 60, in padding *\n if N_FRAMES < self.INPUT_SIZE:\n\n TypeError: '<' not supported between instances of 'NoneType' and 'int'\nEven I'm facing the same issue, anyone who knows how to resolve it….plz help
", "rawMarkdown": "Even I'm facing the same issue, anyone who knows how to resolve it....plz help", "votes": null }, { "id": "3389962", "postDate": "01/12/2026 09:23:43", "content": "good work !!!
", "rawMarkdown": "good work !!!", "votes": null } ], "comments": [ { "id": 2241780, "author_name": "uttammittal02", "author_url": "", "post_date": "05/01/2023 20:09:02", "content": "Even I'm facing the same issue, anyone who knows how to resolve it….plz help
", "votes": null, "replies": [] }, { "id": 3389962, "author_name": "shivanshgoyal123", "author_url": "", "post_date": "01/12/2026 09:23:43", "content": "good work !!!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2241741": "```\nclass FeatureGen(tf.Module):\n def __init__(self):\n super(FeatureGen, self).__init__()\n self.meta_data = metaData\n self.landMark = [0, 468, 489, 522, 543]\n self.INPUT_SIZE = 32\n \n def fill_nans(self,X, l, r):\n z = X[:, l : r]\n nan_idx, _ = tf.unique(tf.where(tf.reduce_all(tf.math.is_nan(z), axis=1))[:, 0])\n not_nan_idx = tf.sets.difference(tf.reshape(tf.range(0, z.shape[0], dtype = tf.dtypes.int64), (1, -1)), tf.reshape(nan_idx, (1, -1))).values\n if (nan_idx.shape[0]==None):\n return z\n if (nan_idx.shape[0] * 4 > not_nan_idx.shape[0]):\n indices = tf.where(tf.math.is_nan(z))\n z = tf.tensor_scatter_nd_update(\n z,\n indices,\n tf.zeros((tf.shape(indices)[0]))\n )\n return z\n if nan_idx[0] == 0:\n k,l = not_nan_idx[0],not_nan_idx[1]\n if (k == l):\n print(f\"Uncaught exception, k can't be equal to l : k = {k}, l = {l}\")\n exit()\n d = (z[l] - z[k]) / (l-k)\n a = z[k]\n z[0] = a - d*k\n not_nan_idx=tf.concat([[0], not_nan_idx], axis = 0)\n nan_idx=nan_idx[1:]\n if (nan_idx.shape[0] == 0):\n return z\n if nan_idx[-1] == z.shape[0] -1 :\n k,l = not_nan_idx[-1],not_nan_idx[-2]\n if (k == l):\n print(f\"Uncaught exception, k can't be equal to l : k = {k}, l = {l}\")\n exit()\n d = (z[k] - z[l]) / (k-l)\n a = z[k]\n z[-1] = a + d*(z.shape[0] -1 - k)\n not_nan_idx=tf.concat([not_nan_idx,[z.shape[0]-1]], axis = 0)\n nan_idx = nan_idx[:-1]\n for i in nan_idx:\n k,l = not_nan_idx[not_nan_idx