06_dip.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d59ca4fc",
   "metadata": {},
   "outputs": [],
   "source": [
    "# default_exp dips"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "be9eabff",
   "metadata": {},
   "outputs": [],
   "source": [
    "# exports\n",
    "import os\n",
    "import pickle\n",
    "from pathlib import Path\n",
    "import numpy as np\n",
    "import torch\n",
    "import llamass.core\n",
    "import llamass.transforms\n",
    "from einops import repeat, rearrange\n",
    "from scipy.spatial.transform import Rotation as R"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "479414af",
   "metadata": {},
   "source": [
    "# DIP\n",
    "\n",
    "> Load the prototype version of AMASS."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3bba3a54",
   "metadata": {},
   "source": [
    "A prototypical version of AMASS can be found in the downloads of [the DIP project](https://dip.is.tuebingen.mpg.de/). This is used in some papers so being able to train on it is useful.\n",
    "\n",
    "Sidenote: The official documentation for the DIPS dataset says the angles stored in `data[\"poses\"]` in it's archives are stored as axis-angle vectors, but when I looked at the code to open this dataset in the SPL repository [here][spl] I was surprised to see they converted it from a rotation matrix to other formalisms. So, I opened it up myself and tested if the data stored was a rotation matrix or not and it was! However, that was even more confusing because there were only 15 joint angles and I knew that SMPL is 24 joint angles but it turns out they didn't include all of them, which you can see if you look at the list of `SMPL_MAJOR_JOINTS` in the [SPL forward kinematics model][splfk].\n",
    "\n",
    "To do:\n",
    "\n",
    "* ~~Visualize an example after applying the SMPL forward kinematics module to verify it's loaded correctly~~\n",
    "\n",
    "[spl]: https://github.com/eth-ait/spl/blob/master/preprocessing/preprocess_dip.py#L183-L192\n",
    "[splfk]: https://github.com/eth-ait/spl/blob/master/visualization/fk.py#L31"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "087ae486",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Contains 975 in a 135 dim vector containing 15 joint angles\n",
      "Example rotation matrix equals identity when multiplied by it's transpose\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[1.00000000e+00, 3.27052040e-17, 4.41648625e-17],\n",
       "       [3.27052040e-17, 1.00000000e+00, 1.40562284e-17],\n",
       "       [4.41648625e-17, 1.40562284e-17, 1.00000000e+00]])"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# spl\n",
    "dips_path = Path(\"/nobackup/gngdb/Synthetic_60FPS/\")\n",
    "\n",
    "# open a random pickle file in the DIPS dataset\n",
    "example_file = None\n",
    "for dirpath, dirnames, filenames in os.walk(dips_path):\n",
    "    dirpath = Path(dirpath)\n",
    "    for filename in filenames:\n",
    "        filename = Path(filename)\n",
    "        if not filename.is_dir() and filename.suffix == \".pkl\":\n",
    "            example_file = dirpath / filename\n",
    "    if example_file is not None:\n",
    "        break\n",
    "\n",
    "def pkl_load(pkl_file):\n",
    "    try:\n",
    "        with open(pkl_file, \"rb\") as f:\n",
    "            return pickle.load(f)\n",
    "    except UnicodeDecodeError:\n",
    "        with open(pkl_file, \"rb\") as f:\n",
    "            u = pickle._Unpickler(f)\n",
    "            u.encoding = \"latin1\"\n",
    "            return u.load()\n",
    "\n",
    "cdata = pkl_load(example_file)\n",
    "\n",
    "n = len(cdata['poses'])\n",
    "d = len(cdata['poses'][0])\n",
    "print(f\"Contains {n} in a {d} dim vector containing {d//9} joint angles\")\n",
    "\n",
    "r = np.array(cdata['poses'][0]).reshape(-1,3,3)[0]\n",
    "assert np.allclose(r.dot(r.T), np.eye(3))\n",
    "print(\"Example rotation matrix equals identity when multiplied by it's transpose\")\n",
    "r.dot(r.T)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "04bf8f93",
   "metadata": {},
   "source": [
    "## Sparse to Full\n",
    "\n",
    "> The data is provided with only 15 major joints, the remaining joints must be padded."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "30cd7c2d",
   "metadata": {},
   "source": [
    "Reference implementation from [SPL](https://github.com/eth-ait/spl/blob/master/common/conversions.py#L211):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5746714c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# spl\n",
    "\"\"\"\n",
    "SPL: training and evaluation of neural networks with a structured prediction layer.\n",
    "Copyright (C) 2019 ETH Zurich, Emre Aksan, Manuel Kaufmann\n",
    "This program is free software: you can redistribute it and/or modify\n",
    "it under the terms of the GNU General Public License as published by\n",
    "the Free Software Foundation, either version 3 of the License, or\n",
    "(at your option) any later version.\n",
    "This program is distributed in the hope that it will be useful,\n",
    "but WITHOUT ANY WARRANTY; without even the implied warranty of\n",
    "MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the\n",
    "GNU General Public License for more details.\n",
    "You should have received a copy of the GNU General Public License\n",
    "along with this program.  If not, see <https://www.gnu.org/licenses/>.\n",
    "\"\"\"\n",
    "def _sparse_to_full(joint_angles_sparse, sparse_joints_idxs, tot_nr_joints, rep=\"rotmat\"):\n",
    "    \"\"\"\n",
    "    Pad the given sparse joint angles with identity elements to retrieve a full skeleton with `tot_nr_joints`\n",
    "    many joints.\n",
    "    Args:\n",
    "        joint_angles_sparse: An np array of shape (N, len(sparse_joints_idxs) * dof)\n",
    "          or (N, len(sparse_joints_idxs), dof)\n",
    "        sparse_joints_idxs: A list of joint indices pointing into the full skeleton given by range(0, tot_nr_joints)\n",
    "        tot_nr_jonts: Total number of joints in the full skeleton.\n",
    "        rep: Which representation is used, rotmat or quat\n",
    "    Returns:\n",
    "        The padded joint angles as an array of shape (N, tot_nr_joints*dof)\n",
    "    \"\"\"\n",
    "    joint_idxs = sparse_joints_idxs\n",
    "    assert rep in [\"rotmat\", \"quat\", \"aa\"]\n",
    "    dof = 9 if rep == \"rotmat\" else 4 if rep == \"quat\" else 3\n",
    "    n_sparse_joints = len(sparse_joints_idxs)\n",
    "    angles_sparse = np.reshape(joint_angles_sparse, [-1, n_sparse_joints, dof])\n",
    "\n",
    "    # fill in the missing indices with the identity element\n",
    "    smpl_full = np.zeros(shape=[angles_sparse.shape[0], tot_nr_joints, dof])  # (N, tot_nr_joints, dof)\n",
    "    if rep == \"quat\":\n",
    "        smpl_full[..., 0] = 1.0\n",
    "    elif rep == \"rotmat\":\n",
    "        smpl_full[..., 0] = 1.0\n",
    "        smpl_full[..., 4] = 1.0\n",
    "        smpl_full[..., 8] = 1.0\n",
    "    else:\n",
    "        pass  # nothing to do for angle-axis\n",
    "\n",
    "    smpl_full[:, joint_idxs] = angles_sparse\n",
    "    smpl_full = np.reshape(smpl_full, [-1, tot_nr_joints * dof])\n",
    "    return smpl_full"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b35c214a",
   "metadata": {},
   "source": [
    "Converting this to PyTorch:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "acbf7083",
   "metadata": {},
   "outputs": [],
   "source": [
    "# exports\n",
    "def sparse_to_full(joint_angles_sparse, sparse_joints_idxs, tot_nr_joints, rep=\"rotmat\"):\n",
    "    \"\"\"\n",
    "    Pad the given sparse joint angles with identity elements to retrieve a full skeleton with `tot_nr_joints`\n",
    "    many joints.\n",
    "    Args:\n",
    "        joint_angles_sparse: Tensor of shape (N, len(sparse_joints_idxs) * dof)\n",
    "          or (N, len(sparse_joints_idxs), dof)\n",
    "        sparse_joints_idxs: A list of joint indices pointing into the full skeleton given by range(0, tot_nr_joints)\n",
    "        tot_nr_jonts: Total number of joints in the full skeleton.\n",
    "        rep: Which representation is used, rotmat or quat\n",
    "    Returns:\n",
    "        The padded joint angles as an array of shape (N, tot_nr_joints*dof)\n",
    "    \"\"\"\n",
    "    device = joint_angles_sparse.device\n",
    "    joint_idxs = sparse_joints_idxs\n",
    "    joint_idx_mapping = {j:i for i,j in enumerate(joint_idxs)}\n",
    "    assert rep in [\"rotmat\", \"quat\", \"aa\"]\n",
    "    dof = 9 if rep == \"rotmat\" else 4 if rep == \"quat\" else 3\n",
    "    n_sparse_joints = len(sparse_joints_idxs)\n",
    "    angles_sparse = joint_angles_sparse.view(-1, n_sparse_joints, dof)\n",
    "\n",
    "    \n",
    "    \n",
    "    # fill in the missing indices with the identity element\n",
    "    N = angles_sparse.size(0)\n",
    "    #smpl_full = torch.zeros((N, tot_nr_joints, dof)).to(device)\n",
    "    if rep == \"quat\":\n",
    "        smpl_full = torch.tensor([1.0, 0., 0., 0.]).to(device)\n",
    "        #smpl_full[..., 0] = 1.0\n",
    "    elif rep == \"rotmat\":\n",
    "        smpl_full = torch.eye(3).view(-1).to(device)\n",
    "        #smpl_full[..., 0] = 1.0\n",
    "        #smpl_full[..., 4] = 1.0\n",
    "        #smpl_full[..., 8] = 1.0\n",
    "    else:\n",
    "        smpl_full = torch.zeros(3).to(device)\n",
    "    \n",
    "    # repeat these tensors along the N axis\n",
    "    smpl_full = repeat(smpl_full, 'd -> N () d', N=N)\n",
    "    \n",
    "    # make a list of tensors for each joint\n",
    "    joint_tensors = []\n",
    "    for j in range(tot_nr_joints):\n",
    "        if j in joint_idxs:\n",
    "            k = joint_idx_mapping[j]\n",
    "            joint_tensors.append(angles_sparse[:, [k]])\n",
    "        else:\n",
    "            joint_tensors.append(smpl_full)\n",
    "\n",
    "    smpl_full =  torch.cat(joint_tensors, 1)\n",
    "    smpl_full = smpl_full.view(-1, tot_nr_joints*dof)\n",
    "    return smpl_full"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7ee533cc",
   "metadata": {},
   "outputs": [],
   "source": [
    "# spl\n",
    "fk = llamass.transforms.SMPL_ForwardKinematics()\n",
    "sparse = torch.tensor(np.array(cdata['poses'][0])).float()\n",
    "_full = _sparse_to_full(sparse.numpy(), fk.major_joints, fk.n_joints, rep=\"rotmat\")\n",
    "full = sparse_to_full(sparse, fk.major_joints, fk.n_joints, rep=\"rotmat\")\n",
    "assert np.allclose(_full, full.numpy())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "00851fc8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# spl\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "positions = fk.from_rotmat(full)\n",
    "\n",
    "def plot_pose(positions, skeleton=None, parents=None, save_to=None):\n",
    "    if parents:\n",
    "        skeleton = [(i,j) for i, j in enumerate(parents)][1:]\n",
    "    assert skeleton is not None\n",
    "    fig, axes = plt.subplots(1, 3, figsize=(10,6))\n",
    "\n",
    "    for d, ax in enumerate(axes):\n",
    "        dims_to_plot = [i for i in range(3) if i != d]\n",
    "        joints = positions\n",
    "        j = joints[:, dims_to_plot]\n",
    "        ax.scatter(*j.T, color=\"b\", s=0.5)\n",
    "        for bone in skeleton:\n",
    "            if set(bone) <= set(range(positions.shape[0])):\n",
    "                a = j[bone[0]]\n",
    "                b = j[bone[1]]\n",
    "                x, y = list(zip(a, b))\n",
    "                ax.plot(x, y, color=\"r\", alpha=0.5)\n",
    "        ax.axes.xaxis.set_ticklabels([])\n",
    "        ax.axes.yaxis.set_ticklabels([])\n",
    "        ax.set_aspect('equal', adjustable='box')\n",
    "    if save_to is not None:\n",
    "        plt.tight_layout()\n",
    "        plt.savefig(save_to)\n",
    "        plt.close()\n",
    "    else:\n",
    "        plt.show()\n",
    "\n",
    "plot_pose(positions[0], parents=fk.parents)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f1210e44",
   "metadata": {},
   "source": [
    "Creating new Forward Kinematics model that incorporates this:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8e4d20c5",
   "metadata": {},
   "outputs": [],
   "source": [
    "# exports\n",
    "class SMPL_ForwardKinematics_Sparse(llamass.transforms.SMPL_ForwardKinematics):\n",
    "    def from_rotmat(self, joint_angles):\n",
    "        mj, nj = self.major_joints, self.n_joints\n",
    "        return super().from_rotmat(sparse_to_full(joint_angles, mj, nj, rep=\"rotmat\"))\n",
    "        \n",
    "    def from_aa(self, joint_angles):\n",
    "        mj, nj = self.major_joints, self.n_joints\n",
    "        return super().from_aa(sparse_to_full(joint_angles, mj, nj, rep=\"aa\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c6d78bcd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# spl\n",
    "fk = SMPL_ForwardKinematics_Sparse()\n",
    "plot_pose(fk.from_rotmat(sparse)[0], parents=fk.parents)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6e24b0b2",
   "metadata": {},
   "source": [
    "# Converting to npz\n",
    "\n",
    "> AMASS proper is stored as npz files, after converting this dataset to that representation, it will take up less space and be easy to load with the AMASS Dataset class."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "016df40e",
   "metadata": {},
   "source": [
    "I've downloaded the txt files containing the lists of files to go into each of training, test and validation splits from [here](https://github.com/eth-ait/spl/tree/master/preprocessing).\n",
    "\n",
    "To do:\n",
    "\n",
    "* Console utility to process the data into npz files like the AMASS dataset\n",
    "    * ~~Get list of all the files~~\n",
    "    * Split the list of files according to the `train.txt`...\n",
    "    * Process each archive in parallel with joblib, saving to new location\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ab8c8e71",
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture\n",
    "%%bash\n",
    "wget -O training_fnames.txt https://raw.githubusercontent.com/eth-ait/spl/master/preprocessing/training_fnames.txt\n",
    "wget -O test_fnames.txt https://raw.githubusercontent.com/eth-ait/spl/master/preprocessing/test_fnames.txt\n",
    "wget -O validation_fnames.txt https://raw.githubusercontent.com/eth-ait/spl/master/preprocessing/validation_fnames.txt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8b69efb7",
   "metadata": {},
   "outputs": [],
   "source": [
    "# exports\n",
    "def iter_pkl_in(dip_dir):\n",
    "    # open a random pickle file in the DIPS dataset\n",
    "    example_file = None\n",
    "    for dirpath, dirnames, filenames in os.walk(dips_path):\n",
    "        dirpath = Path(dirpath)\n",
    "        for filename in filenames:\n",
    "            filename = Path(filename)\n",
    "            if not filename.is_dir() and filename.suffix == \".pkl\":\n",
    "                yield dirpath/filename"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b9210da4",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/nobackup/gngdb/Synthetic_60FPS/AMASS_ACCAD/Female1General_c3dA10_SB__SB2__SB_lie_SB_to_SB_crouch_dynamics.pkl\n"
     ]
    }
   ],
   "source": [
    "# spl\n",
    "for pkl_path in iter_pkl_in(dips_path):\n",
    "    print(pkl_path)\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e86385dd",
   "metadata": {},
   "outputs": [],
   "source": [
    "# spl\n",
    "def load_txt(fpath):\n",
    "    with open(fpath) as f:\n",
    "        return [Path(p.rstrip()) for p in f.readlines()]\n",
    "split_paths = {}\n",
    "split_paths[\"train\"] = load_txt(\"training_fnames.txt\")\n",
    "split_paths[\"validation\"] = load_txt(\"validation_fnames.txt\")\n",
    "split_paths[\"test\"] = load_txt(\"test_fnames.txt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9cd00086",
   "metadata": {},
   "outputs": [],
   "source": [
    "# spl\n",
    "pkey = lambda p: (p.parts[-2].split(\"_\")[-1], p.name)\n",
    "split_fnames = {pkey(p):n for n,v in split_paths.items() for p in v}\n",
    "separated_paths = {k:[] for k in split_paths}\n",
    "allocated, unallocated = [], []\n",
    "for pkl_path in iter_pkl_in(dips_path):\n",
    "    try:\n",
    "        split = split_fnames[pkey(pkl_path)]\n",
    "        assert pkey(pkl_path) not in allocated\n",
    "        allocated.append(pkey(pkl_path))\n",
    "        separated_paths[split].append(pkl_path)\n",
    "    except KeyError:\n",
    "        unallocated.append(pkl_path)\n",
    "n = len(set(p for k,v in separated_paths.items() for p in v))\n",
    "prescribed = set(k for k in split_fnames)\n",
    "assert n == len(prescribed)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "641f1a3f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# spl\n",
    "from tqdm.auto import tqdm\n",
    "def process_to_npz():\n",
    "    with tqdm(total=len(prescribed)) as pbar:\n",
    "        for split, pkl_paths in separated_paths.items():\n",
    "            for pkl_path in pkl_paths:\n",
    "                dip_path = pkl_path.parents[2]\n",
    "                dest_dir =  dip_path / split / pkl_path.parts[-2].split(\"_\")[-1]\n",
    "                dest_path = dest_dir / (pkl_path.parts[-1].split(\".\")[0] + \".npz\")\n",
    "                dest_dir.mkdir(exist_ok=True, parents=True)\n",
    "                with open(pkl_path, \"rb\") as f:\n",
    "                    cdata = pickle.load(f, encoding='latin1')\n",
    "                np.savez(dest_path, poses=np.array(cdata[\"poses\"]))\n",
    "                pbar.update(1)\n",
    "# process_to_npz()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "21c7e87e",
   "metadata": {},
   "source": [
    "Trying to open this new processed dataset using the AMASS Dataset class:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d238aef3",
   "metadata": {},
   "outputs": [],
   "source": [
    "# spl\n",
    "dip_path = Path(\"/nobackup/gngdb/dips/train/\")\n",
    "dataset = llamass.core.AMASS(\n",
    "    dip_path,\n",
    "    clip_length=144,\n",
    "    overlapping=False,\n",
    "    transform=torch.tensor,\n",
    "    data_keys=(\"poses\",),\n",
    "    strict=False\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "424280bc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'poses': tensor([[ 0.9903, -0.1298,  0.0501,  ..., -0.8230, -0.5596,  0.0976],\n",
      "        [ 0.9904, -0.1307,  0.0457,  ..., -0.8153, -0.5775,  0.0428],\n",
      "        [ 0.9904, -0.1320,  0.0400,  ..., -0.8051, -0.5930, -0.0100],\n",
      "        ...,\n",
      "        [ 0.9402, -0.2332,  0.2483,  ..., -0.7959, -0.5205, -0.3091],\n",
      "        [ 0.9378, -0.2353,  0.2554,  ..., -0.7994, -0.5183, -0.3038],\n",
      "        [ 0.9377, -0.2353,  0.2557,  ..., -0.7965, -0.5208, -0.3071]])}\n",
      "poses torch.Size([144, 135])\n"
     ]
    }
   ],
   "source": [
    "# spl\n",
    "for example in dataset:\n",
    "    print(example)\n",
    "    for k, v in example.items():\n",
    "        print(k, v.size())\n",
    "    break"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d5b06f2e",
   "metadata": {},
   "source": [
    "Is it possible to load all of DIP into memory?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "40695452",
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_tensor_dataset():\n",
    "    dips_examples = []\n",
    "    for example in tqdm(dataset):\n",
    "        dips_examples.append(example)\n",
    "    dip = torch.stack([e[\"poses\"] for e in dips_examples])\n",
    "    print(dip.size(), f\" number of poses={dip.size(0)*dip.size(1)}\")\n",
    "    torch.save(dip, dip_path/\"tensor_dataset.pt\")\n",
    "# build_tensor_dataset()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "71be2ccc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.0G\t/nobackup/gngdb/dips/train/tensor_dataset.pt\r\n"
     ]
    }
   ],
   "source": [
    "!du -hs /nobackup/gngdb/dips/train/tensor_dataset.pt"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "65f55e41",
   "metadata": {},
   "source": [
    "# Adding Additional Training Data\n",
    "\n",
    "> AMASS contains extra data, can this be added to assist during training?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7fca1805",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}