sosaa-data/experiments/ood-synthesis.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "6c32f936",
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib as mpl\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "import numpy as np\n",
    "\n",
    "import sklearn\n",
    "\n",
    "from sklearn.neighbors import KernelDensity\n",
    "from sklearn.neural_network import MLPClassifier\n",
    "\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "3111cd4c",
   "metadata": {
    "papermill": {
     "duration": 0.015913,
     "end_time": "2022-11-29T15:48:54.753987",
     "exception": false,
     "start_time": "2022-11-29T15:48:54.738074",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "import colorcet\n",
    "\n",
    "conf_cmap = \"cet_CET_L20\"\n",
    "diff_cmap = \"cet_CET_L20\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "d0a7c561",
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_toy_line(seed):\n",
    "    rng = np.random.default_rng(seed)\n",
    "\n",
    "    X1 = np.concatenate([np.linspace(0, 4, 5000), np.linspace(6, 10, 5000)])\n",
    "    X2 = 1 + X1 * 0.5 + rng.normal(loc=0.0, scale=0.1, size=10000)\n",
    "    X1 += rng.normal(loc=0.0, scale=0.1, size=10000)\n",
    "    Y = X1*np.sqrt(1.25)\n",
    "    \n",
    "    return np.stack([X1, X2], axis=1), Y\n",
    "\n",
    "def generate_toy_line_test():\n",
    "    Xl1T, Xl2T = np.mgrid[-2:12:0.01, -1:8:0.01]\n",
    "    Xl1T = Xl1T.flatten()\n",
    "    Xl2T = Xl2T.flatten()\n",
    "\n",
    "    return np.stack([Xl1T, Xl2T], axis=1)\n",
    "\n",
    "def optimal_toy_line_reconstruction(Xl):\n",
    "    return tf.constant([[0.0, 1.0]]) + tf.constant([[1.0, 0.5]]) * (\n",
    "        tf.linalg.matmul((Xl - tf.constant([0.0, 1.0])), tf.constant([[1.0], [0.5]])) /\n",
    "        tf.linalg.matmul(tf.constant([[1.0, 0.5]]), tf.constant([[1.0], [0.5]]))\n",
    "    )\n",
    "\n",
    "Xl, Yl = generate_toy_line(42)\n",
    "XlT = generate_toy_line_test()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "68f83245",
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_toy_circle(seed):\n",
    "    rng = np.random.default_rng(seed)\n",
    "    \n",
    "    Y = np.sin(np.linspace(0, np.pi * 8.0, 10000))\n",
    "    \n",
    "    X1 = np.sin(np.linspace(0, np.pi * 2.0, 10000)) * (5 + Y)\n",
    "    X2 = np.cos(np.linspace(0, np.pi * 2.0, 10000)) * (5 + Y)\n",
    "    \n",
    "    X1 += rng.normal(loc=0.0, scale=0.1, size=10000)\n",
    "    X2 += rng.normal(loc=0.0, scale=0.1, size=10000)\n",
    "    \n",
    "    return np.stack([X1, X2], axis=1), Y\n",
    "\n",
    "def generate_toy_circle_test():\n",
    "    Xl1T, Xl2T = np.mgrid[-7.5:7.5:0.01, -7.5:7.5:0.01]\n",
    "    Xl1T = Xl1T.flatten()\n",
    "    Xl2T = Xl2T.flatten()\n",
    "\n",
    "    return np.stack([Xl1T, Xl2T], axis=1)\n",
    "\n",
    "def optimal_toy_circle_reconstruction(Xc):\n",
    "    angle = tf.math.atan2(Xc[:,0], Xc[:,1])\n",
    "    \n",
    "    Y = tf.math.sin(angle * 4.0)\n",
    "    \n",
    "    X1 = tf.math.sin(angle) * (5.0 + Y)\n",
    "    X2 = tf.math.cos(angle) * (5.0 + Y)\n",
    "    \n",
    "    return tf.stack([X1, X2], axis=1)\n",
    "\n",
    "Xc, Yc = generate_toy_circle(42)\n",
    "XcT = generate_toy_circle_test()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "6d3d4652",
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_toy_haystack(seed):\n",
    "    rng = np.random.default_rng(seed)\n",
    "    \n",
    "    n = 10\n",
    "    a = 2\n",
    "    \n",
    "    # https://stats.stackexchange.com/a/124554\n",
    "    A = np.matrix([rng.normal(size=n) + rng.normal(size=1)*a for i in range(n)])\n",
    "    A = A*np.transpose(A)\n",
    "    D_half = np.diag(np.diag(A)**(-0.5))\n",
    "    C = D_half*A*D_half\n",
    "    covs = C\n",
    "            \n",
    "    Cn = rng.choice([-2, -1, 0, 0, 1, 2], size=10)\n",
    "    \n",
    "    Xn = rng.multivariate_normal(np.zeros(shape=10), covs, size=10000)\n",
    "    \n",
    "    Xn[:,3] = -0.42\n",
    "    \n",
    "    Yn = np.dot(Xn, Cn)\n",
    "    \n",
    "    return Xn, Yn\n",
    "\n",
    "def generate_toy_haystack_test(seed):\n",
    "    rng = np.random.default_rng(seed)\n",
    "    \n",
    "    n = 10\n",
    "    a = 2\n",
    "    \n",
    "    # https://stats.stackexchange.com/a/124554\n",
    "    A = np.matrix([rng.normal(size=n) + rng.normal(size=1)*a for i in range(n)])\n",
    "    A = A*np.transpose(A)\n",
    "    D_half = np.diag(np.diag(A)**(-0.5))\n",
    "    C = D_half*A*D_half\n",
    "    covs = C\n",
    "            \n",
    "    Cn = rng.choice([-2, -1, 0, 0, 1, 2], size=10)\n",
    "    \n",
    "    rng = np.random.default_rng(rng.integers(0, 2**30))\n",
    "    \n",
    "    XnT = rng.multivariate_normal(np.zeros(shape=10), covs, size=10000)\n",
    "    \n",
    "    XnT[:,3] = (rng.random(size=10000) - 0.5) - 0.42\n",
    "    \n",
    "    return XnT\n",
    "\n",
    "def optimal_toy_haystack_reconstruction(Xh):\n",
    "    return (\n",
    "        Xh * tf.constant([[1,1,1,0,1,1,1,1,1,1]], dtype=tf.float32) -\n",
    "        tf.constant([[0,0,0,0.42,0,0,0,0,0,0]], dtype=tf.float32)\n",
    "    )\n",
    "\n",
    "Xh, Yh = generate_toy_haystack(42)\n",
    "XhT = generate_toy_haystack_test(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "db6d78dc",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_line_toy_confidence(Xl, XlO, XlT, ClT, short, title):\n",
    "    fig, ax = plt.subplots(1, 1, figsize=(6, 4))\n",
    "\n",
    "    ax.imshow(\n",
    "        ClT.reshape(np.mgrid[-2:12:0.01, -1:8:0.01][0].shape).T, cmap=conf_cmap, vmin=0.0, vmax=1.2,\n",
    "        extent=[-2,12,-1,8], origin=\"lower\", interpolation=\"bicubic\", rasterized=True,\n",
    "    )\n",
    "\n",
    "    ax.set_xlim(-2, 12)\n",
    "    ax.set_ylim(-1, 8)\n",
    "    \n",
    "    ax.scatter(XlO[::25,0], XlO[::25,1], c='white', s=2, rasterized=True)\n",
    "    ax.scatter(Xl[::500,0], Xl[::500,1], c='white', marker='x', lw=3, s=48)\n",
    "    ax.scatter(Xl[::500,0], Xl[::500,1], c='black', marker='x')\n",
    "    \n",
    "    ax.text(\n",
    "        0.5, 0.95, title, ha=\"center\", va=\"top\",\n",
    "        size=20, c=\"white\", bbox=dict(facecolor='black', alpha=0.25, edgecolor='white'),\n",
    "        transform=ax.transAxes,\n",
    "    )\n",
    "\n",
    "    ax.axis('off')\n",
    "\n",
    "    plt.savefig(f\"ood-line-{short}.pdf\", dpi=100, transparent=True, bbox_inches='tight')\n",
    "    # plt.show()\n",
    "    plt.close(fig)\n",
    "\n",
    "def plot_line_toy_difference(Xl, XlO, XlOW, XlT, ClT, short, title):\n",
    "    fig, ax = plt.subplots(1, 1, figsize=(6, 4))\n",
    "    \n",
    "    vlim = np.amax(np.abs(ClT))\n",
    "\n",
    "    ax.imshow(\n",
    "        ClT.reshape(np.mgrid[-2:12:0.01, -1:8:0.01][0].shape).T, cmap=diff_cmap, vmin=-vlim,\n",
    "        vmax=vlim, extent=[-2,12,-1,8], origin=\"lower\", interpolation=\"bicubic\", rasterized=True,\n",
    "    )\n",
    "\n",
    "    ax.set_xlim(-2, 12)\n",
    "    ax.set_ylim(-1, 8)\n",
    "    \n",
    "    ax.scatter(XlO[::5,0], XlO[::5,1], c='white', alpha=1.0-XlOW[::5], s=2, rasterized=True)\n",
    "    \n",
    "    ax.text(\n",
    "        0.5, 0.95, title, ha=\"center\", va=\"top\",\n",
    "        size=20, c=\"black\", bbox=dict(facecolor='white', alpha=0.5, edgecolor='black'),\n",
    "        transform=ax.transAxes,\n",
    "    )\n",
    "\n",
    "    ax.axis('off')\n",
    "\n",
    "    plt.savefig(f\"ood-line-{short}.pdf\", dpi=100, transparent=True, bbox_inches='tight')\n",
    "    # plt.show()\n",
    "    plt.close(fig)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "bc35e725",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_circle_toy_confidence(Xl, XlO, XlT, ClT, short):\n",
    "    fig, ax = plt.subplots(1, 1, figsize=(6, 4))\n",
    "\n",
    "    ax.imshow(\n",
    "        ClT.reshape(np.mgrid[-7.5:7.5:0.01, -7.5:7.5:0.01][0].shape).T, cmap=conf_cmap, vmin=0.0, vmax=1.2,\n",
    "        extent=[-7.5,7.5,-7.5,7.5], origin=\"lower\", interpolation=\"bicubic\", rasterized=True,\n",
    "    )\n",
    "\n",
    "    ax.set_xlim(-7.5, 7.5)\n",
    "    ax.set_ylim(-7.5, 7.5)\n",
    "    \n",
    "    ax.scatter(XlO[::25,0], XlO[::25,1], c='white', s=4, rasterized=True)\n",
    "    ax.scatter(Xl[::250,0], Xl[::250,1], c='white', marker='x', lw=3, s=48)\n",
    "    ax.scatter(Xl[::250,0], Xl[::250,1], c='black', marker='x')\n",
    "\n",
    "    ax.axis('off')\n",
    "\n",
    "    plt.savefig(f\"ood-circle-{short}.pdf\", dpi=100, transparent=True, bbox_inches='tight')\n",
    "    # plt.show()\n",
    "    plt.close(fig)\n",
    "\n",
    "def plot_circle_toy_difference(Xl, XlO, XlOW, XlT, ClT, short):\n",
    "    fig, ax = plt.subplots(1, 1, figsize=(6, 4))\n",
    "    \n",
    "    vlim = np.amax(np.abs(ClT))\n",
    "\n",
    "    ax.imshow(\n",
    "        ClT.reshape(np.mgrid[-7.5:7.5:0.01, -7.5:7.5:0.01][0].shape).T, cmap=diff_cmap, vmin=-vlim,\n",
    "        vmax=vlim, extent=[-7.5,7.5,-7.5,7.5], origin=\"lower\", interpolation=\"bicubic\", rasterized=True,\n",
    "    )\n",
    "\n",
    "    ax.set_xlim(-7.5, 7.5)\n",
    "    ax.set_ylim(-7.5, 7.5)\n",
    "    \n",
    "    ax.scatter(XlO[::5,0], XlO[::5,1], c='white', alpha=1.0-XlOW[::5], s=2, rasterized=True)\n",
    "\n",
    "    ax.axis('off')\n",
    "\n",
    "    plt.savefig(f\"ood-circle-{short}.pdf\", dpi=100, transparent=True, bbox_inches='tight')\n",
    "    # plt.show()\n",
    "    plt.close(fig)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "3101cc1e",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_haystack_toy_confidence(XlT, ClT, short):\n",
    "    fig, ax = plt.subplots(1, 1, figsize=(6, 4))\n",
    "    \n",
    "    ax.spines['top'].set_visible(False)\n",
    "    ax.spines['right'].set_visible(False)\n",
    "    ax.spines['bottom'].set_visible(False)\n",
    "    ax.spines['left'].set_visible(False)\n",
    "    \n",
    "    ax.set_xticks([-0.42-0.4, -0.42, -0.42+0.4])\n",
    "    ax.set_xticklabels([r\"$-0.4 \\sigma$\", \"const\", \"$+0.4 \\sigma$\"])\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "    \n",
    "    ax.axvline(-0.42, c='white', lw=5, zorder=-1)\n",
    "    \n",
    "    ax.scatter(XlT[:, 3], ClT, c='white', s=6, rasterized=True)\n",
    "    ax.scatter(XlT[:, 3], ClT, c='black', s=1, rasterized=True)\n",
    "    \n",
    "    ax.imshow(\n",
    "        [[x/100, x/100] for x in range(100, -1, -1)], cmap=conf_cmap, interpolation='bicubic',\n",
    "        vmin=0.0, vmax=1.2, extent=[-0.93, 0.09, -0.01, 1.01], zorder=-2,\n",
    "    )\n",
    "    \n",
    "    class TruncatedColormap(mpl.colors.Colormap):\n",
    "        def __init__(self, cmap):\n",
    "            self.cmap = cmap\n",
    "            self.N = cmap.N\n",
    "        \n",
    "        def __call__(self, X, alpha=None, bytes=False):\n",
    "            return self.cmap.__call__(X/1.2, alpha=alpha, bytes=bytes)\n",
    "    \n",
    "    cb = fig.colorbar(mpl.cm.ScalarMappable(\n",
    "        norm=None, cmap=TruncatedColormap(mpl.cm.get_cmap(conf_cmap)),\n",
    "    ), ax=ax, ticks=[0.1, 0.9])\n",
    "    cb.set_label(\"confidence level $c$\", labelpad=-13)\n",
    "    cb.outline.set_visible(False)\n",
    "    \n",
    "    plt.savefig(f\"ood-haystack-{short}.pdf\", dpi=100, transparent=True, bbox_inches='tight')\n",
    "    # plt.show()\n",
    "    plt.close(fig)\n",
    "\n",
    "def plot_haystack_toy_difference(XlT, ClT, short):\n",
    "    fig, ax = plt.subplots(1, 1, figsize=(6, 4))\n",
    "    \n",
    "    ax.spines['top'].set_visible(False)\n",
    "    ax.spines['right'].set_visible(False)\n",
    "    ax.spines['bottom'].set_visible(False)\n",
    "    ax.spines['left'].set_visible(False)\n",
    "    \n",
    "    ax.set_xticks([-0.42-0.4, -0.42, -0.42+0.4])\n",
    "    ax.set_xticklabels([r\"$-0.4 \\sigma$\", \"const\", \"$+0.4 \\sigma$\"])\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "    \n",
    "    vlim = np.amax(np.abs(ClT))\n",
    "    \n",
    "    ax.axvline(-0.42, c='white', lw=5, zorder=-1)\n",
    "    \n",
    "    ax.scatter(XlT[:, 3], (ClT / vlim) * 0.5 + 0.5, c='white', s=6, rasterized=True)\n",
    "    ax.scatter(XlT[:, 3], (ClT / vlim) * 0.5 + 0.5, c='black', s=1, rasterized=True)\n",
    "    \n",
    "    ax.imshow(\n",
    "        [[x/100, x/100] for x in range(100, -1, -1)], cmap=diff_cmap, interpolation='bicubic',\n",
    "        extent=[-0.93, 0.09, -0.01, 1.01], zorder=-2,\n",
    "    )\n",
    "    \n",
    "    cb = fig.colorbar(mpl.cm.ScalarMappable(\n",
    "        norm=None, cmap=diff_cmap,\n",
    "    ), ax=ax, ticks=[0.1, 0.9])\n",
    "    cb.ax.set_yticklabels([\"$<$ \", \"$>$ \"])\n",
    "    cb.ax.tick_params(pad=3)\n",
    "    cb.set_label(\"change in confidence level $c$\", labelpad=-13)\n",
    "    cb.outline.set_visible(False)\n",
    "    \n",
    "    plt.savefig(f\"ood-haystack-{short}.pdf\", dpi=100, transparent=True, bbox_inches='tight')\n",
    "    # plt.show()\n",
    "    plt.close(fig)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "0c05f719",
   "metadata": {},
   "outputs": [],
   "source": [
    "Xl_ood_u = np.random.default_rng(24).uniform(np.amin(Xl, axis=0) - 1, np.amax(Xl, axis=0) + 1, size=(10000, 2))\n",
    "\n",
    "NN_l_u = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xl, Xl_ood_u]), np.concatenate([np.ones(shape=len(Xl)), np.zeros(shape=len(Xl_ood_u))])\n",
    ")\n",
    "\n",
    "plot_line_toy_confidence(\n",
    "    Xl, Xl_ood_u, XlT, NN_l_u.predict_proba(XlT)[:,1], \"uniform\", \"OOD $\\sim$ Uniform\"\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "76038d7a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-03-24 22:21:57.524819: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  SSE4.1 SSE4.2 AVX AVX2 FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
      "2023-03-24 22:21:57.527168: I tensorflow/core/common_runtime/process_util.cc:146] Creating new thread pool with default inter op setting: 2. Tune using inter_op_parallelism_threads for best performance.\n"
     ]
    }
   ],
   "source": [
    "kde_l = KernelDensity(bandwidth=0.05, kernel=\"gaussian\").fit(Xl)\n",
    "XlS = kde_l.sample(10000, random_state=42)\n",
    "\n",
    "xl_id = tf.constant(XlS, dtype=tf.float32)\n",
    "\n",
    "with tf.GradientTape() as tape:\n",
    "    tape.watch(xl_id)\n",
    "\n",
    "    xl_pred_id = optimal_toy_line_reconstruction(xl_id)\n",
    "    xl_mse_id = (xl_pred_id - xl_id) ** 2\n",
    "\n",
    "adv_grad_l = tape.gradient(xl_mse_id, xl_id)\n",
    "\n",
    "Xl_ood_g = xl_id + tf.math.sign(adv_grad_l) * 1.0\n",
    "\n",
    "NN_l_g = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xl, Xl_ood_g]), np.concatenate([np.ones(shape=len(Xl)), np.zeros(shape=len(Xl_ood_g))])\n",
    ")\n",
    "\n",
    "plot_line_toy_confidence(\n",
    "    Xl, Xl_ood_g, XlT, NN_l_g.predict_proba(XlT)[:,1], \"fgsm\", r\"OOD $\\sim$ FGSM$(\\epsilon = 1)$\"\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "5cc7c856",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.15 0.9324014771268805\n",
      "0.1575 0.9409286372330858\n",
      "0.16537500000000002 0.9484458707448783\n",
      "0.17364375000000004 0.9551706335674071\n",
      "0.17190731250000005 0.9533282886932147\n",
      "=> 0.172\n"
     ]
    }
   ],
   "source": [
    "t = 0.15\n",
    "\n",
    "for _ in range(10):\n",
    "    Xl_ood_gt = xl_id + tf.math.sign(adv_grad_l) * t\n",
    "\n",
    "    NN_l_gt = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "        np.concatenate([Xl, Xl_ood_gt]), np.concatenate([np.ones(shape=len(Xl)), np.zeros(shape=len(Xl_ood_gt))])\n",
    "    )\n",
    "    \n",
    "    mean_id_conf = np.mean(NN_l_gt.predict_proba(Xl)[:,1])\n",
    "    \n",
    "    print(t, mean_id_conf)\n",
    "    \n",
    "    if mean_id_conf < 0.95:\n",
    "        t *= 1.05\n",
    "    elif mean_id_conf >= 0.955:\n",
    "        t *= 0.99\n",
    "    else:\n",
    "        break\n",
    "        \n",
    "print(f\"=> {t:.03}\")\n",
    "\n",
    "plot_line_toy_confidence(\n",
    "    Xl, Xl_ood_gt, XlT, NN_l_gt.predict_proba(XlT)[:,1], \"poking-fgsm\", f\"OOD $\\sim$ FGSM$(\\epsilon = t_{{poke}})$\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "924ecf56",
   "metadata": {},
   "outputs": [],
   "source": [
    "Xl_ood_gu = xl_id + tf.math.sign(adv_grad_l) * tf.math.abs(\n",
    "    tf.random.uniform(minval=0.0, maxval=1.0, shape=(len(xl_id), 1), seed=242)\n",
    ")\n",
    "\n",
    "NN_l_gu = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xl, Xl_ood_gu]), np.concatenate([np.ones(shape=len(Xl)), np.zeros(shape=len(Xl_ood_gu))])\n",
    ")\n",
    "\n",
    "plot_line_toy_confidence(\n",
    "    Xl, Xl_ood_gu, XlT, NN_l_gu.predict_proba(XlT)[:,1], \"uniform-fgsm\", \"OOD $\\sim$ FGSM$(\\epsilon \\sim$ U$(0, 1))$\"\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "992d3dc2",
   "metadata": {},
   "outputs": [],
   "source": [
    "ElT = np.array(optimal_toy_line_reconstruction(Xl) - Xl)\n",
    "ElOgu = np.array(optimal_toy_line_reconstruction(Xl_ood_gu) - Xl_ood_gu)\n",
    "\n",
    "kde_lT = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(ElT)\n",
    "kde_lOgu = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(ElOgu)\n",
    "kde_lOguD = np.exp(kde_lOgu.score_samples(ElOgu))\n",
    "\n",
    "kde_lTOgu = np.exp(kde_lT.score_samples(ElOgu))\n",
    "\n",
    "ilgu = np.argmax(kde_lTOgu)\n",
    "flgu = kde_lOguD[ilgu] / kde_lTOgu[ilgu]\n",
    "\n",
    "sample_weight_lgu = np.concatenate([\n",
    "    np.ones(shape=len(Xl)),\n",
    "    1.0 - ((kde_lTOgu * flgu / np.amax(kde_lOguD)) / np.maximum(\n",
    "        kde_lTOgu * flgu / np.amax(kde_lOguD), kde_lOguD / np.amax(kde_lOguD)\n",
    "    )),\n",
    "])\n",
    "\n",
    "include_sample_lgu = np.random.default_rng(424).random(size=len(sample_weight_lgu)) < sample_weight_lgu\n",
    "\n",
    "NN_l_guw = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xl, Xl_ood_gu])[include_sample_lgu],\n",
    "    np.concatenate([np.ones(shape=len(Xl)), np.zeros(shape=len(Xl_ood_gu))])[include_sample_lgu],\n",
    ")\n",
    "\n",
    "plot_line_toy_confidence(\n",
    "    Xl, Xl_ood_gu[include_sample_lgu[len(Xl):]], XlT, NN_l_guw.predict_proba(XlT)[:,1],\n",
    "    \"uniform-fgsm-weighted\", \"OOD-Weighting for\\nOOD $\\sim$ FGSM$(\\epsilon \\sim$ U$(0, 1))$\",\n",
    ")\n",
    "\n",
    "plot_line_toy_difference(\n",
    "    Xl, Xl_ood_gu, sample_weight_lgu[len(Xl):], XlT,\n",
    "    NN_l_guw.predict_proba(XlT)[:,1] - NN_l_gu.predict_proba(XlT)[:,1],\n",
    "    \"uniform-fgsm-weighted-difference\", \"Impact of OOD-Weighting for\\nOOD $\\sim$ FGSM$(\\epsilon \\sim$ U$(0, 1))$\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "57914cd8",
   "metadata": {},
   "outputs": [],
   "source": [
    "ElT = np.array(optimal_toy_line_reconstruction(Xl) - Xl)\n",
    "ElO = np.array(optimal_toy_line_reconstruction(Xl_ood_u) - Xl_ood_u)\n",
    "\n",
    "kde_lT = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(ElT)\n",
    "kde_lO = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(ElO)\n",
    "kde_lOD = np.exp(kde_lO.score_samples(ElO))\n",
    "\n",
    "kde_lTO = np.exp(kde_lT.score_samples(ElO))\n",
    "\n",
    "il = np.argmax(kde_lTO)\n",
    "fl = kde_lOD[il] / kde_lTO[il]\n",
    "\n",
    "sample_weight_l = np.concatenate([\n",
    "    np.ones(shape=len(Xl)),\n",
    "    1.0 - ((kde_lTO * fl / np.amax(kde_lOD)) / np.maximum(\n",
    "        kde_lTO * fl / np.amax(kde_lOD), kde_lOD / np.amax(kde_lOD)\n",
    "    )),\n",
    "])\n",
    "\n",
    "include_sample_l = np.random.default_rng(424).random(size=len(sample_weight_l)) < sample_weight_l\n",
    "\n",
    "NN_l_w = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xl, Xl_ood_u])[include_sample_l],\n",
    "    np.concatenate([np.ones(shape=len(Xl)), np.zeros(shape=len(Xl_ood_u))])[include_sample_l],\n",
    ")\n",
    "\n",
    "plot_line_toy_confidence(\n",
    "    Xl, Xl_ood_u[include_sample_l[len(Xl):]], XlT, NN_l_w.predict_proba(XlT)[:,1],\n",
    "    \"uniform-weighted\", \"OOD-Weighting for\\nOOD $\\sim$ Uniform\",\n",
    ")\n",
    "\n",
    "plot_line_toy_difference(\n",
    "    Xl, Xl_ood_u, sample_weight_l[len(Xl):], XlT,\n",
    "    NN_l_w.predict_proba(XlT)[:,1] - NN_l_u.predict_proba(XlT)[:,1],\n",
    "    \"uniform-weighted-difference\", \"Impact of OOD-Weighting for\\nOOD $\\sim$ Uniform\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "b757161d",
   "metadata": {},
   "outputs": [],
   "source": [
    "Xc_ood_u = np.random.default_rng(24).uniform(np.amin(Xc, axis=0) - 1, np.amax(Xc, axis=0) + 1, size=(10000, 2))\n",
    "\n",
    "NN_c_u = MLPClassifier(activation=\"relu\", max_iter=500, random_state=42).fit(\n",
    "    np.concatenate([Xc, Xc_ood_u]), np.concatenate([np.ones(shape=len(Xc)), np.zeros(shape=len(Xc_ood_u))])\n",
    ")\n",
    "\n",
    "plot_circle_toy_confidence(\n",
    "    Xc, Xc_ood_u, XcT, NN_c_u.predict_proba(XcT)[:,1], \"uniform\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "c2f53112",
   "metadata": {},
   "outputs": [],
   "source": [
    "kde_c = KernelDensity(bandwidth=0.05, kernel=\"gaussian\").fit(Xc)\n",
    "XcS = kde_c.sample(10000, random_state=42)\n",
    "\n",
    "xc_id = tf.constant(XcS, dtype=tf.float32)\n",
    "\n",
    "with tf.GradientTape() as tape:\n",
    "    tape.watch(xc_id)\n",
    "\n",
    "    xc_pred_id = optimal_toy_circle_reconstruction(xc_id)\n",
    "    xc_mse_id = (xc_pred_id - xc_id) ** 2\n",
    "\n",
    "adv_grad_c = tape.gradient(xc_mse_id, xc_id)\n",
    "\n",
    "Xc_ood_g = xc_id + tf.math.sign(adv_grad_c) * 1.0\n",
    "\n",
    "NN_c_g = MLPClassifier(activation=\"relu\", max_iter=500, random_state=42).fit(\n",
    "    np.concatenate([Xc, Xc_ood_g]), np.concatenate([np.ones(shape=len(Xc)), np.zeros(shape=len(Xc_ood_g))])\n",
    ")\n",
    "\n",
    "plot_circle_toy_confidence(\n",
    "    Xc, Xc_ood_g, XcT, NN_c_g.predict_proba(XcT)[:,1], \"fgsm\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "58201dc2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.46 0.9590524053261795\n",
      "0.4554 0.9612162780229968\n",
      "0.450846 0.9518878958697329\n",
      "=> 0.451\n"
     ]
    }
   ],
   "source": [
    "t = 0.46\n",
    "\n",
    "for _ in range(10):\n",
    "    Xc_ood_gt = xc_id + tf.math.sign(adv_grad_c) * t\n",
    "\n",
    "    NN_c_gt = MLPClassifier(activation=\"relu\", max_iter=1500, random_state=42).fit(\n",
    "        np.concatenate([Xc, Xc_ood_gt]), np.concatenate([np.ones(shape=len(Xc)), np.zeros(shape=len(Xc_ood_gt))])\n",
    "    )\n",
    "    \n",
    "    mean_id_conf = np.mean(NN_c_gt.predict_proba(Xc)[:,1])\n",
    "    \n",
    "    print(t, mean_id_conf)\n",
    "    \n",
    "    if mean_id_conf < 0.95:\n",
    "        t *= 1.05\n",
    "    elif mean_id_conf >= 0.955:\n",
    "        t *= 0.99\n",
    "    else:\n",
    "        break\n",
    "        \n",
    "print(f\"=> {t:.03}\")\n",
    "\n",
    "plot_circle_toy_confidence(\n",
    "    Xc, Xc_ood_gt, XcT, NN_c_gt.predict_proba(XcT)[:,1], \"poking-fgsm\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "ff6f016b",
   "metadata": {},
   "outputs": [],
   "source": [
    "Xc_ood_gu = xc_id + tf.math.sign(adv_grad_c) * tf.math.abs(\n",
    "    tf.random.uniform(minval=0.0, maxval=1.0, shape=(len(xc_id), 1), seed=242)\n",
    ")\n",
    "\n",
    "NN_c_gu = MLPClassifier(activation=\"relu\", max_iter=1000, random_state=42).fit(\n",
    "    np.concatenate([Xc, Xc_ood_gu]), np.concatenate([np.ones(shape=len(Xc)), np.zeros(shape=len(Xc_ood_gu))])\n",
    ")\n",
    "\n",
    "plot_circle_toy_confidence(\n",
    "    Xc, Xc_ood_gu, XcT, NN_c_gu.predict_proba(XcT)[:,1], \"uniform-fgsm\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "0995f374",
   "metadata": {},
   "outputs": [],
   "source": [
    "EcT = np.array(optimal_toy_circle_reconstruction(Xc) - Xc)\n",
    "EcOgu = np.array(optimal_toy_circle_reconstruction(Xc_ood_gu) - Xc_ood_gu)\n",
    "\n",
    "kde_cT = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(EcT)\n",
    "kde_cOgu = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(EcOgu)\n",
    "kde_cOguD = np.exp(kde_cOgu.score_samples(EcOgu))\n",
    "\n",
    "kde_cTOgu = np.exp(kde_cT.score_samples(EcOgu))\n",
    "\n",
    "icgu = np.argmax(kde_cTOgu)\n",
    "fcgu = kde_cOguD[icgu] / kde_cTOgu[icgu]\n",
    "\n",
    "sample_weight_cgu = np.concatenate([\n",
    "    np.ones(shape=len(Xc)),\n",
    "    1.0 - ((kde_cTOgu * fcgu / np.amax(kde_cOguD)) / np.maximum(\n",
    "        kde_cTOgu * fcgu / np.amax(kde_cOguD), kde_cOguD / np.amax(kde_cOguD)\n",
    "    )),\n",
    "])\n",
    "\n",
    "include_sample_cgu = np.random.default_rng(424).random(size=len(sample_weight_cgu)) < sample_weight_cgu\n",
    "\n",
    "NN_c_guw = MLPClassifier(activation=\"relu\", max_iter=1000, random_state=42).fit(\n",
    "    np.concatenate([Xc, Xc_ood_gu])[include_sample_cgu],\n",
    "    np.concatenate([np.ones(shape=len(Xc)), np.zeros(shape=len(Xc_ood_gu))])[include_sample_cgu],\n",
    ")\n",
    "\n",
    "plot_circle_toy_confidence(\n",
    "    Xc, Xc_ood_gu[include_sample_cgu[len(Xc):]], XcT, NN_c_guw.predict_proba(XcT)[:,1],\n",
    "    \"uniform-fgsm-weighted\",\n",
    ")\n",
    "\n",
    "plot_circle_toy_difference(\n",
    "    Xc, Xc_ood_gu, sample_weight_cgu[len(Xc):], XcT,\n",
    "    NN_c_guw.predict_proba(XcT)[:,1] - NN_c_gu.predict_proba(XcT)[:,1],\n",
    "    \"uniform-fgsm-weighted-difference\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "af1205b6",
   "metadata": {},
   "outputs": [],
   "source": [
    "EcT = np.array(optimal_toy_circle_reconstruction(Xc) - Xc)\n",
    "EcO = np.array(optimal_toy_circle_reconstruction(Xc_ood_u) - Xc_ood_u)\n",
    "\n",
    "kde_cT = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(EcT)\n",
    "kde_cO = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(EcO)\n",
    "kde_cOD = np.exp(kde_cO.score_samples(EcO))\n",
    "\n",
    "kde_cTO = np.exp(kde_cT.score_samples(EcO))\n",
    "\n",
    "ic = np.argmax(kde_cTO)\n",
    "fc = kde_cOD[ic] / kde_cTO[ic]\n",
    "\n",
    "sample_weight_c = np.concatenate([\n",
    "    np.ones(shape=len(Xc)),\n",
    "    1.0 - ((kde_cTO * fc / np.amax(kde_cOD)) / np.maximum(\n",
    "        kde_cTO * fc / np.amax(kde_cOD), kde_cOD / np.amax(kde_cOD)\n",
    "    )),\n",
    "])\n",
    "\n",
    "include_sample_c = np.random.default_rng(424).random(size=len(sample_weight_c)) < sample_weight_c\n",
    "\n",
    "NN_c_w = MLPClassifier(activation=\"relu\", max_iter=500, random_state=42).fit(\n",
    "    np.concatenate([Xc, Xc_ood_u])[include_sample_c],\n",
    "    np.concatenate([np.ones(shape=len(Xc)), np.zeros(shape=len(Xc_ood_u))])[include_sample_c],\n",
    ")\n",
    "\n",
    "plot_circle_toy_confidence(\n",
    "    Xc, Xc_ood_u[include_sample_c[len(Xc):]], XcT, NN_c_w.predict_proba(XcT)[:,1],\n",
    "    \"uniform-weighted\",\n",
    ")\n",
    "\n",
    "plot_circle_toy_difference(\n",
    "    Xc, Xc_ood_u, sample_weight_c[len(Xc):], XcT,\n",
    "    NN_c_w.predict_proba(XcT)[:,1] - NN_c_u.predict_proba(XcT)[:,1],\n",
    "    \"uniform-weighted-difference\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "100bc1a5",
   "metadata": {},
   "outputs": [],
   "source": [
    "Xh_ood_u = np.random.default_rng(24).uniform(np.amin(Xh, axis=0) - 1, np.amax(Xh, axis=0) + 1, size=(10000, 10))\n",
    "\n",
    "NN_h_u = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xh, Xh_ood_u]), np.concatenate([np.ones(shape=len(Xh)), np.zeros(shape=len(Xh_ood_u))])\n",
    ")\n",
    "\n",
    "plot_haystack_toy_confidence(\n",
    "    XhT, NN_h_u.predict_proba(XhT)[:,1], \"uniform\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "14349c2f",
   "metadata": {},
   "outputs": [],
   "source": [
    "kde_h = KernelDensity(bandwidth=0.05, kernel=\"gaussian\").fit(Xh)\n",
    "XhS = kde_h.sample(10000, random_state=42)\n",
    "\n",
    "xh_id = tf.constant(XhS, dtype=tf.float32)\n",
    "\n",
    "with tf.GradientTape() as tape:\n",
    "    tape.watch(xh_id)\n",
    "\n",
    "    xh_pred_id = optimal_toy_haystack_reconstruction(xh_id)\n",
    "    xh_mse_id = (xh_pred_id - xh_id) ** 2\n",
    "\n",
    "adv_grad_h = tape.gradient(xh_mse_id, xh_id)\n",
    "\n",
    "Xh_ood_g = xh_id + tf.math.sign(adv_grad_h) * 1.0\n",
    "\n",
    "NN_h_g = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xh, Xh_ood_g]), np.concatenate([np.ones(shape=len(Xh)), np.zeros(shape=len(Xh_ood_g))])\n",
    ")\n",
    "\n",
    "plot_haystack_toy_confidence(\n",
    "    XhT, NN_h_g.predict_proba(XhT)[:,1], \"fgsm\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "e2eff255",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.018 0.998051672185284\n",
      "0.01782 0.517419055964483\n",
      "0.018711 0.9966334477095538\n",
      "0.018523889999999998 0.9980953197803528\n",
      "0.0183386511 0.517321444909457\n",
      "0.019255583655 0.9974442166289939\n",
      "0.01906302781845 0.9971731933284816\n",
      "0.018872397540265498 0.9976521947544364\n",
      "0.018683673564862842 0.9977951474129669\n",
      "0.018496836829214214 0.9972955754889088\n",
      "=> 0.0183\n"
     ]
    }
   ],
   "source": [
    "t = 0.018\n",
    "\n",
    "for _ in range(10):\n",
    "    Xh_ood_gt = xh_id + tf.math.sign(adv_grad_h) * t\n",
    "\n",
    "    NN_h_gt = MLPClassifier(activation=\"relu\", max_iter=500, random_state=42).fit(\n",
    "        np.concatenate([Xh, Xh_ood_gt]), np.concatenate([np.ones(shape=len(Xh)), np.zeros(shape=len(Xh_ood_gt))])\n",
    "    )\n",
    "    \n",
    "    mean_id_conf = np.mean(NN_h_gt.predict_proba(Xh)[:,1])\n",
    "    \n",
    "    print(t, mean_id_conf)\n",
    "    \n",
    "    if mean_id_conf < 0.95:\n",
    "        t *= 1.05\n",
    "    elif mean_id_conf >= 0.955:\n",
    "        t *= 0.99\n",
    "    else:\n",
    "        break\n",
    "        \n",
    "print(f\"=> {t:.03}\")\n",
    "\n",
    "plot_haystack_toy_confidence(\n",
    "    XhT, NN_h_gt.predict_proba(XhT)[:,1], \"poking-fgsm\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "d2bcaebd",
   "metadata": {},
   "outputs": [],
   "source": [
    "Xh_ood_gu = xh_id + tf.math.sign(adv_grad_h) * tf.math.abs(\n",
    "    tf.random.uniform(minval=0.0, maxval=1.0, shape=(len(xh_id), 1), seed=242)\n",
    ")\n",
    "\n",
    "NN_h_gu = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xh, Xh_ood_gu]), np.concatenate([np.ones(shape=len(Xh)), np.zeros(shape=len(Xh_ood_gu))])\n",
    ")\n",
    "\n",
    "plot_haystack_toy_confidence(\n",
    "    XhT, NN_h_gu.predict_proba(XhT)[:,1], \"uniform-fgsm\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "5878608a",
   "metadata": {},
   "outputs": [],
   "source": [
    "EhT = np.array(optimal_toy_haystack_reconstruction(Xh) - Xh)\n",
    "EhOgu = np.array(optimal_toy_haystack_reconstruction(Xh_ood_gu) - Xh_ood_gu)\n",
    "\n",
    "kde_hT = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(EhT)\n",
    "kde_hOgu = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(EhOgu)\n",
    "kde_hOguD = np.exp(kde_hOgu.score_samples(EhOgu))\n",
    "\n",
    "kde_hTOgu = np.exp(kde_hT.score_samples(EhOgu))\n",
    "\n",
    "ihgu = np.argmax(kde_hTOgu)\n",
    "fhgu = kde_hOguD[ihgu] / kde_hTOgu[ihgu]\n",
    "\n",
    "sample_weight_hgu = np.concatenate([\n",
    "    np.ones(shape=len(Xh)),\n",
    "    1.0 - ((kde_hTOgu * fhgu / np.amax(kde_hOguD)) / np.maximum(\n",
    "        kde_hTOgu * fhgu / np.amax(kde_hOguD), kde_hOguD / np.amax(kde_hOguD)\n",
    "    )),\n",
    "])\n",
    "\n",
    "include_sample_hgu = np.random.default_rng(424).random(size=len(sample_weight_hgu)) < sample_weight_hgu\n",
    "\n",
    "NN_h_guw = MLPClassifier(activation=\"relu\", max_iter=500, random_state=42).fit(\n",
    "    np.concatenate([Xh, Xh_ood_gu])[include_sample_hgu],\n",
    "    np.concatenate([np.ones(shape=len(Xh)), np.zeros(shape=len(Xh_ood_gu))])[include_sample_hgu],\n",
    ")\n",
    "\n",
    "plot_haystack_toy_confidence(\n",
    "    XhT, NN_h_guw.predict_proba(XhT)[:,1],\n",
    "    \"uniform-fgsm-weighted\",\n",
    ")\n",
    "\n",
    "plot_haystack_toy_difference(\n",
    "    XhT, NN_h_guw.predict_proba(XhT)[:,1] - NN_h_gu.predict_proba(XhT)[:,1],\n",
    "    \"uniform-fgsm-weighted-difference\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "2b66fecd",
   "metadata": {},
   "outputs": [],
   "source": [
    "EhT = np.array(optimal_toy_haystack_reconstruction(Xh) - Xh)\n",
    "EhO = np.array(optimal_toy_haystack_reconstruction(Xh_ood_u) - Xh_ood_u)\n",
    "\n",
    "kde_hT = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(EhT)\n",
    "kde_hO = KernelDensity(bandwidth=0.1, kernel=\"linear\").fit(EhO)\n",
    "kde_hOD = np.exp(kde_hO.score_samples(EhO))\n",
    "\n",
    "kde_hTO = np.exp(kde_hT.score_samples(EhO))\n",
    "\n",
    "ih = np.argmax(kde_hTO)\n",
    "fh = kde_hOD[ih] / kde_hTO[ih]\n",
    "\n",
    "sample_weight_h = np.concatenate([\n",
    "    np.ones(shape=len(Xh)),\n",
    "    1.0 - ((kde_hTO * fh / np.amax(kde_hOD)) / np.maximum(\n",
    "        kde_hTO * fh / np.amax(kde_hOD), kde_hOD / np.amax(kde_hOD)\n",
    "    )),\n",
    "])\n",
    "\n",
    "include_sample_h = np.random.default_rng(424).random(size=len(sample_weight_h)) < sample_weight_h\n",
    "\n",
    "NN_h_w = MLPClassifier(activation=\"relu\", random_state=42).fit(\n",
    "    np.concatenate([Xh, Xh_ood_u])[include_sample_h],\n",
    "    np.concatenate([np.ones(shape=len(Xh)), np.zeros(shape=len(Xh_ood_u))])[include_sample_h],\n",
    ")\n",
    "\n",
    "plot_haystack_toy_confidence(\n",
    "    XhT, NN_h_w.predict_proba(XhT)[:,1],\n",
    "    \"uniform-weighted\",\n",
    ")\n",
    "\n",
    "plot_haystack_toy_difference(\n",
    "    XhT, NN_h_w.predict_proba(XhT)[:,1] - NN_h_u.predict_proba(XhT)[:,1],\n",
    "    \"uniform-weighted-difference\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b2cb5e18",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.6"
  },
  "papermill": {
   "default_parameters": {},
   "duration": 165.016529,
   "end_time": "2022-11-29T15:51:28.691885",
   "environment_variables": {},
   "exception": null,
   "input_path": "demo2.ipynb",
   "output_path": "demo2.ipynb",
   "parameters": {},
   "start_time": "2022-11-29T15:48:43.675356",
   "version": "2.3.4"
  },
  "widgets": {
   "application/vnd.jupyter.widget-state+json": {
    "state": {
     "06c1c7a790cb4c79b2b09ae4b836d980": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "ProgressStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "bar_color": null,
       "description_width": ""
      }
     },
     "12f6437e0b014208ad97864661e8cfc3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HTMLView",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_e45a7381d25b435b8be8b9b1171734d4",
       "placeholder": "​",
       "style": "IPY_MODEL_412bb756667b4cfea22a2fb5a770e2f1",
       "tabbable": null,
       "tooltip": null,
       "value": " 100/100 [00:37&lt;00:00,  2.59it/s]"
      }
     },
     "1450e20b86d748c78ed031512bc090a7": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "179d084ad94b4fe498cecd0af0743148": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "1b090a5937c148ceb7a7b4cedb75ca3f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HTMLView",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_4e96ba4f2c8e4401b442a19d291ddb62",
       "placeholder": "​",
       "style": "IPY_MODEL_cc77f3cf079446d1b84a4ca6cb345955",
       "tabbable": null,
       "tooltip": null,
       "value": " 100/100 [00:38&lt;00:00,  2.63it/s]"
      }
     },
     "20bf597ef3df4cb3b40fbcbeb2fb5ed0": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "22fa99f1f46240eabed2ace04291c3d6": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "2c48a14cccce46dfba2928e40f31e796": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "ProgressStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "bar_color": null,
       "description_width": ""
      }
     },
     "2e5a8120cf404898a8e332e5e5896fe8": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "3a138ff615c942c4a4e62db6d577d94e": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HBoxModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HBoxModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HBoxView",
       "box_style": "",
       "children": [
        "IPY_MODEL_eae55362b025496980e7b70db78cb1f5",
        "IPY_MODEL_824b6df432974d56aeec3b395a3a3193",
        "IPY_MODEL_12f6437e0b014208ad97864661e8cfc3"
       ],
       "layout": "IPY_MODEL_563ddec360bb4948ae7bab979ef98873",
       "tabbable": null,
       "tooltip": null
      }
     },
     "412bb756667b4cfea22a2fb5a770e2f1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "background": null,
       "description_width": "",
       "font_size": null,
       "text_color": null
      }
     },
     "429aa48ab7ac4b18a4a4d070e8de85a3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HBoxModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HBoxModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HBoxView",
       "box_style": "",
       "children": [
        "IPY_MODEL_b2ee7b8da9c4455a83e328e8f400377d",
        "IPY_MODEL_bbda44e373a54a2ba8031fd109924cb3",
        "IPY_MODEL_452caa04e5cb4682b26f4d76de941d66"
       ],
       "layout": "IPY_MODEL_5d6cc35355ff4269bf5b814e967ed15b",
       "tabbable": null,
       "tooltip": null
      }
     },
     "4385a7f7ae0e459999bc5e4c4fc9d6e5": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "452caa04e5cb4682b26f4d76de941d66": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HTMLView",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_4b7ddb7ccdd145829c532f8462a78ebd",
       "placeholder": "​",
       "style": "IPY_MODEL_94bc158b17b04f18899dcbc918241186",
       "tabbable": null,
       "tooltip": null,
       "value": " 100/100 [00:38&lt;00:00,  2.59it/s]"
      }
     },
     "4b7ddb7ccdd145829c532f8462a78ebd": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "4bbe3e72c8a447439c4848712afe1288": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HTMLView",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_a2b0d35a24364b59beb59c2516ea4eea",
       "placeholder": "​",
       "style": "IPY_MODEL_da22cf76520e4394b0c67fc3acfeb9c7",
       "tabbable": null,
       "tooltip": null,
       "value": "100%"
      }
     },
     "4cad5af313924729a810e017afec4565": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "background": null,
       "description_width": "",
       "font_size": null,
       "text_color": null
      }
     },
     "4e96ba4f2c8e4401b442a19d291ddb62": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "4f1738d98404462488cdd7db8d8f5429": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "FloatProgressModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "FloatProgressModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "ProgressView",
       "bar_style": "success",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_4385a7f7ae0e459999bc5e4c4fc9d6e5",
       "max": 100,
       "min": 0,
       "orientation": "horizontal",
       "style": "IPY_MODEL_82c6782e38714066a4279d96a7820136",
       "tabbable": null,
       "tooltip": null,
       "value": 100
      }
     },
     "522f65c4d9ae484f9b16b4b9ec88a485": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "background": null,
       "description_width": "",
       "font_size": null,
       "text_color": null
      }
     },
     "563ddec360bb4948ae7bab979ef98873": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "5a27ef10885a4270aee1f8f921a54220": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HBoxModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HBoxModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HBoxView",
       "box_style": "",
       "children": [
        "IPY_MODEL_d8b0944bea314a7f9756ebb74304cf34",
        "IPY_MODEL_4f1738d98404462488cdd7db8d8f5429",
        "IPY_MODEL_1b090a5937c148ceb7a7b4cedb75ca3f"
       ],
       "layout": "IPY_MODEL_1450e20b86d748c78ed031512bc090a7",
       "tabbable": null,
       "tooltip": null
      }
     },
     "5d6cc35355ff4269bf5b814e967ed15b": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "824b6df432974d56aeec3b395a3a3193": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "FloatProgressModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "FloatProgressModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "ProgressView",
       "bar_style": "success",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_af2b2b0b2ada4f4e95b35379b3303bd6",
       "max": 100,
       "min": 0,
       "orientation": "horizontal",
       "style": "IPY_MODEL_f74404aedd684d09b5845ecdcb032bc6",
       "tabbable": null,
       "tooltip": null,
       "value": 100
      }
     },
     "82c6782e38714066a4279d96a7820136": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "ProgressStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "bar_color": null,
       "description_width": ""
      }
     },
     "926dab2dd69e4270af1d87cfb446dc89": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "94bc158b17b04f18899dcbc918241186": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "background": null,
       "description_width": "",
       "font_size": null,
       "text_color": null
      }
     },
     "a2b0d35a24364b59beb59c2516ea4eea": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "a47e42c8c01242278528c9ef837bc4a1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HBoxModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HBoxModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HBoxView",
       "box_style": "",
       "children": [
        "IPY_MODEL_4bbe3e72c8a447439c4848712afe1288",
        "IPY_MODEL_ebef2a4014ab47e9ad60f365ad338a6f",
        "IPY_MODEL_ca41aa579e4f44a78bc266369e1b7ab1"
       ],
       "layout": "IPY_MODEL_2e5a8120cf404898a8e332e5e5896fe8",
       "tabbable": null,
       "tooltip": null
      }
     },
     "af2b2b0b2ada4f4e95b35379b3303bd6": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "b2ee7b8da9c4455a83e328e8f400377d": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HTMLView",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_edfcae3d4945472e8eb943912f8e8b38",
       "placeholder": "​",
       "style": "IPY_MODEL_d0241aa138ed4827a465056537538ab5",
       "tabbable": null,
       "tooltip": null,
       "value": "100%"
      }
     },
     "bbda44e373a54a2ba8031fd109924cb3": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "FloatProgressModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "FloatProgressModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "ProgressView",
       "bar_style": "success",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_20bf597ef3df4cb3b40fbcbeb2fb5ed0",
       "max": 100,
       "min": 0,
       "orientation": "horizontal",
       "style": "IPY_MODEL_2c48a14cccce46dfba2928e40f31e796",
       "tabbable": null,
       "tooltip": null,
       "value": 100
      }
     },
     "c8f902e37ecd4ad9bc723afefec2fe92": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "background": null,
       "description_width": "",
       "font_size": null,
       "text_color": null
      }
     },
     "ca41aa579e4f44a78bc266369e1b7ab1": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HTMLView",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_926dab2dd69e4270af1d87cfb446dc89",
       "placeholder": "​",
       "style": "IPY_MODEL_522f65c4d9ae484f9b16b4b9ec88a485",
       "tabbable": null,
       "tooltip": null,
       "value": " 100/100 [00:38&lt;00:00,  2.54it/s]"
      }
     },
     "cc77f3cf079446d1b84a4ca6cb345955": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "background": null,
       "description_width": "",
       "font_size": null,
       "text_color": null
      }
     },
     "d0241aa138ed4827a465056537538ab5": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "background": null,
       "description_width": "",
       "font_size": null,
       "text_color": null
      }
     },
     "d8b0944bea314a7f9756ebb74304cf34": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HTMLView",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_22fa99f1f46240eabed2ace04291c3d6",
       "placeholder": "​",
       "style": "IPY_MODEL_c8f902e37ecd4ad9bc723afefec2fe92",
       "tabbable": null,
       "tooltip": null,
       "value": "100%"
      }
     },
     "da22cf76520e4394b0c67fc3acfeb9c7": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "background": null,
       "description_width": "",
       "font_size": null,
       "text_color": null
      }
     },
     "e45a7381d25b435b8be8b9b1171734d4": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "eae55362b025496980e7b70db78cb1f5": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "HTMLModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "HTMLModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "HTMLView",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_179d084ad94b4fe498cecd0af0743148",
       "placeholder": "​",
       "style": "IPY_MODEL_4cad5af313924729a810e017afec4565",
       "tabbable": null,
       "tooltip": null,
       "value": "100%"
      }
     },
     "ebef2a4014ab47e9ad60f365ad338a6f": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "FloatProgressModel",
      "state": {
       "_dom_classes": [],
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "FloatProgressModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/controls",
       "_view_module_version": "2.0.0",
       "_view_name": "ProgressView",
       "bar_style": "success",
       "description": "",
       "description_allow_html": false,
       "layout": "IPY_MODEL_fc0cac1bb77c4acfb263d4018425acb9",
       "max": 100,
       "min": 0,
       "orientation": "horizontal",
       "style": "IPY_MODEL_06c1c7a790cb4c79b2b09ae4b836d980",
       "tabbable": null,
       "tooltip": null,
       "value": 100
      }
     },
     "edfcae3d4945472e8eb943912f8e8b38": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     },
     "f74404aedd684d09b5845ecdcb032bc6": {
      "model_module": "@jupyter-widgets/controls",
      "model_module_version": "2.0.0",
      "model_name": "ProgressStyleModel",
      "state": {
       "_model_module": "@jupyter-widgets/controls",
       "_model_module_version": "2.0.0",
       "_model_name": "ProgressStyleModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "StyleView",
       "bar_color": null,
       "description_width": ""
      }
     },
     "fc0cac1bb77c4acfb263d4018425acb9": {
      "model_module": "@jupyter-widgets/base",
      "model_module_version": "2.0.0",
      "model_name": "LayoutModel",
      "state": {
       "_model_module": "@jupyter-widgets/base",
       "_model_module_version": "2.0.0",
       "_model_name": "LayoutModel",
       "_view_count": null,
       "_view_module": "@jupyter-widgets/base",
       "_view_module_version": "2.0.0",
       "_view_name": "LayoutView",
       "align_content": null,
       "align_items": null,
       "align_self": null,
       "border_bottom": null,
       "border_left": null,
       "border_right": null,
       "border_top": null,
       "bottom": null,
       "display": null,
       "flex": null,
       "flex_flow": null,
       "grid_area": null,
       "grid_auto_columns": null,
       "grid_auto_flow": null,
       "grid_auto_rows": null,
       "grid_column": null,
       "grid_gap": null,
       "grid_row": null,
       "grid_template_areas": null,
       "grid_template_columns": null,
       "grid_template_rows": null,
       "height": null,
       "justify_content": null,
       "justify_items": null,
       "left": null,
       "margin": null,
       "max_height": null,
       "max_width": null,
       "min_height": null,
       "min_width": null,
       "object_fit": null,
       "object_position": null,
       "order": null,
       "overflow": null,
       "padding": null,
       "right": null,
       "top": null,
       "visibility": null,
       "width": null
      }
     }
    },
    "version_major": 2,
    "version_minor": 0
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}