visualize_regression.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "2e94ef27-6398-41aa-a089-6426fd4285b9",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import os\n",
    "import numpy as np\n",
    "\n",
    "from copy import deepcopy\n",
    "\n",
    "from HyperNetClasses import IdentityNet, NormalVariationalNet\n",
    "\n",
    "from RegressionDataset import SineDataset, LineDataset\n",
    "\n",
    "from _utils import train_val_split_regression\n",
    "\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "d1b4a80b-e653-48cf-b314-d0da90903bec",
   "metadata": {},
   "outputs": [],
   "source": [
    "# import the meta-learning algorithm class of interest\n",
    "from Maml import Maml\n",
    "from Vampire2 import Vampire2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "b2db3121-3a1a-4c2b-8693-ccdc45befd1c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# the index to concatenate to the name of a saved check-point, which will be loaded for visualization\n",
    "resume_maml = 300\n",
    "resume_vampire2 = 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "dfe5638a-c87d-43e0-908c-6768b96ee86d",
   "metadata": {},
   "outputs": [],
   "source": [
    "config = {}\n",
    "\n",
    "config['logdir'] = '.'\n",
    "\n",
    "config['datasource'] = ['SineLine']\n",
    "config['network_architecture'] = 'FcNet'\n",
    "config['num_ways'] = 1\n",
    "config['k_shot'] = 5\n",
    "config['v_shot'] = 10\n",
    "config['num_models'] = 16\n",
    "config['KL_weight'] = 1e-5\n",
    "\n",
    "config['inner_lr'] = 0.001\n",
    "config['num_inner_updates'] = 5\n",
    "config['meta_lr'] = 1e-3\n",
    "\n",
    "config['train_flag'] = False\n",
    "config['num_episodes'] = 1\n",
    "\n",
    "config['device'] = torch.device('cuda:0' if torch.cuda.is_available() else torch.device('cpu'))\n",
    "\n",
    "config['loss_function'] = torch.nn.MSELoss()\n",
    "config['train_val_split_function'] = train_val_split_regression\n",
    "\n",
    "config['first_order'] = True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "4106ab4e-e8f5-4074-a429-ab5b54efaabc",
   "metadata": {},
   "outputs": [],
   "source": [
    "config_maml = deepcopy(config)\n",
    "config_maml['logdir'] = os.path.join(config['logdir'], 'meta_learning', 'maml', config['network_architecture'])\n",
    "maml = Maml(config=config_maml)\n",
    "\n",
    "config_vampire2 = deepcopy(config)\n",
    "config_vampire2['logdir'] = os.path.join(config['logdir'], 'meta_learning', 'vampire2', config['network_architecture'])\n",
    "vampire2 = Vampire2(config=config_vampire2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "891fe3b8-8e57-42b8-9c36-4742aa5d01f4",
   "metadata": {},
   "outputs": [],
   "source": [
    "sine_generator = SineDataset(amplitude_range=[2, 5], phase_range=[0, np.pi], noise_std=0.3, x_range=[-5, 5], num_samples=50)\n",
    "line_generator = LineDataset(slope_range=[-3, 3], intercept_range=[-3, 3], x_range=[-5, 5], num_samples=50, noise_std=0.3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "656faf34-c54d-4cfb-a001-382ac389f512",
   "metadata": {},
   "outputs": [],
   "source": [
    "regression_dataset = torch.utils.data.ConcatDataset(datasets=[sine_generator, line_generator])\n",
    "\n",
    "train_dataloader = torch.utils.data.DataLoader(dataset=regression_dataset)\n",
    "test_dataloader = torch.utils.data.DataLoader(dataset=regression_dataset)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "b38422ee-4c90-4bfd-a66c-948b3a71f51b",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/n10/venv/lib/python3.8/site-packages/torch/nn/modules/lazy.py:178: UserWarning: Lazy modules are a new feature under heavy development so changes to the API or functionality can happen at any moment.\n",
      "  warnings.warn('Lazy modules are a new feature under heavy development '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of parameters of the base network = 1,761.\n",
      "\n",
      "Number of parameters of the base network = 1,761.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "models = {}\n",
    "models['maml'] = maml.load_model(resume_epoch=resume_maml, eps_dataloader=train_dataloader, hyper_net_class=IdentityNet)\n",
    "models['vampire2'] = vampire2.load_model(resume_epoch=resume_vampire2, eps_dataloader=train_dataloader, hyper_net_class=NormalVariationalNet)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "cdf42b96-8c67-474e-bb8e-250a1389a9fa",
   "metadata": {},
   "outputs": [],
   "source": [
    "# set training mode to disable dropout\n",
    "for meta_learning_algorithm in ['maml', 'vampire2']:\n",
    "    models[meta_learning_algorithm]['f_base_net'].eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "17a16287-f83f-4447-9af7-637b58992dfd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAARgAAAEYCAYAAACHjumMAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8rg+JYAAAACXBIWXMAAAsTAAALEwEAmpwYAABR4klEQVR4nO2dd3gc1fWw3ztbtOpdtuQqWe7dlgvuDbBpjk0IEAghCSEQQkhCSSDJL4HgJEBIJSEfmJJQAgkdjMFNxr3IuPcq2bLVi9W2zv3+WHkteXdllV2tyn2fZx/QzJ2ZI3t95pxzTxFSShQKhSIYaKEWQKFQdF2UglEoFEFDKRiFQhE0lIJRKBRBQykYhUIRNIyhFqAlJCUlyf79+4daDIVCcQk7duwokVImX3q8UymY/v37k5OTE2oxFArFJQghcn0dVy6SQqEIGkrBKBSKoKEUjEKhCBpKwSgUiqChFIxCoQgaSsEoFIqgoRSMQqEIGkrBKBSKoKEUjEKhCBpdUsHYz53DUVwcajEUim5Pl1Qwsq4Oe25eqMVQKLo9XVLBADjP5iPt9lCLoVB0a7qsgnFVVeMoKwu1GApFt6bLKhhhMuHIzw+1GApFtybkCkYIYRBC7BRCfBLI+2pRUTgLC5WbpFCEkJArGOAB4GDA7yoE6DrO8oqA31qhUDSPkCoYIURv4FpgaVDuH2ZRbpJCEUJCbcH8GXgE0P0tEELcLYTIEULkFLcwt0WLisJRWIB0ONompUKhaBUhUzBCiOuAIinljqbWSSlfkFJmSSmzkpO9Wn42/QxNA5cLZ3l5W0RVKBStJJQWzFTgBiHEKeAtYI4Q4vVAP0SEWXCcPRvo2yoUimYQMgUjpXxUStlbStkfuAVYI6W8PdDP0aKicJwrQDqdgb61QqG4DKGOwQQdoWngdOJSbpJC0e50CAUjpVwrpbwuWPcXYWE48pWbpFC0Nx1CwQQbLSoK+7lzyk1SKNqZbqFghMEALieuiopQi6JQdCu6hYIBEOYwtZukULQz3UbBaFFR2PPPIl2uUIuiUHQbuo2CcbtJLuUmKRTtSLdRMFDfwuHcuVCLoVB0G7qVgtGionCcOaPcJIWinehWCkYYjUinC1dlZahFUSi6Bd1KwYBykxSK9qTbKRjlJikU7Ue3UzDCaEQ6nLjOnw+1KApFl6fbKRgAjEYcBQWhlkKh6PJ0SwVjiI7Gcfo0UvfbSE+hUASAbqlghNGIbrOp3SSFIsh0SwUDIIwmHEVFoRZDoejSdFsFY4iOxpGXp9wkhSKIdFsFI0wmpM2GrnaTFIqg0W0VDACaAUehcpMUimDRrRWMIToa++nTSClDLYpC0SXp1gpGmM1Iq1W5SQpFkOjWCgYATcPZwomRCoWieYRysqNFCLFNCLFbCLFfCPF4KOTQoqKw5eYpN0mhCAKhtGBswBwp5WhgDDBfCDG5vYXQwsKQdbXo1dXt/WiFossTysmOUkp54V+1qf4TGjNCCOUmKRRBIKQxGCGEQQixCygCVkopt/pYc7cQIkcIkVMcJCWgRcdgz1NukkIRaEKqYKSULinlGKA3MFEIMcLHmheklFlSyqzk5OSgyKGFhaFXV6PX1ATl/gpFd6VD7CJJKSuAbGB+yIQQGs4i5SYpFIEklLtIyUKIuPr/DweuBA6FSh4tOhp7Xq5ykxSKABJKCyYVyBZC7AG2447BfBIqYbSwMPQq5SYpFIHEGKoHSyn3AGODcW+ry4ZV2olq4XXCoOEsKsIQ1dIrFQqFLzpEDCaQVFgrWLj9e/yvdkOLrxVR0dhzlZukUASKLqdg4ixxDI7K4K3aL6jVrS26VgsLQ69RSXcKRaDocgoG4HuDv0WlrOXd8mxsdjv//vgjPlq7luLy8stfLFBJd10YXddZvXo1v/3tbzl27FioxenydEkFM3bAVCaaB/FywceMu/1rvL96Da9++AGjv3oj42+5mR8+9Xv2HDni81otOgb7qVPKTepi5OXl8cQTTzBgwAAefPBBTp8+zZQpU7jqqqt4//33cTqdoRaxS9IlFcz6TZvY+eoR6gw27vr5Tbz/5z/z3p/+TN7nK3jp14/TLzWVhQ/8kJ2HDnpdq4WFodfWKTepC7FkyRLGjh1LYWEh7777Ljt37uT5558nLy+Pb37zm/zhD38gPT2d3bt3h1rULofoTG/qrKwsmZOT0+Sa0tJSpk+fzmM/+QkDEuyMTBuPEMJr3SfrvuCHv/89H/7lL4wcOKjROWdpCZZBgwjLzAyo/Ir25/XXX+eXv/wlmzdvpmfPnn7X/ec//+FnP/sZW7dubXKdwjdCiB1SyqxLj4dsmzpYJCYmsn//fgCq161D2my4wowYReNf9boZM7E7nHzlRz/i4789x7CMDM85LSoaW24e5gEDfConRedg3bp1/OQnPyE7O/uySuPWW2/l6NGjLFy4kLVr1xIeHt5OUnZtuqSLJIRACIE5I4MPSlbztRM/x6rbvNYtnjuX3/3wARY+8EOO5OZ6jrtbONShV1W1p9iKAHL48GFuuukm3nzzTYYPH96sa375y18yYMAAvvWtb6kYXIDokgrmAqaUFPpqSZx1lPBBxTqfa7529dX88u67uenBn2Cz2y+e0ATOwsJ2klTRXKSUfLavoEkFUFJSwrXXXsuSJUuYN29es+8thODll18mNzeXxx8PSf+zLkeXVjBaWBhZfacy1jyAN8o+w6Y7fK674/obyOzbl//3v/9dvDYqGlueagje0cg+XMQ9r+8g+7D/aRDf/OY3ufHGG7nrrrtafH+LxcIHH3zAq6++yocfftgWURV0cQUDENavL9+0zKbYWcEnlf6ze5fc/0Oefe3fnlwZj5ukdpM6DFJKlixz7/wtWXbQp/Jfu3Ythw4d4je/+U2rn9OjRw+WLl3KI488orav20iXVzCG+HiyIoYwIiyD18s+wyV9T3Ickp7OV+ddye9eWuo5JjTV6a4jkX24iHOV7uzsc5VWLytGSsmjjz7KE088gdlsbtOz5s6dS1paGv/+97/bdJ/uTpdXMMJgIKx/Oj8Ov45ne/8Qg/D/Kz/23e/y7qpVHDx5wn1tZBT2vNPtJaqiCS5YL7V2FwC1dpeXFfPxxx9TU1PDrbfe2ubnCSFYsmQJjz/+ODab9waBonl0eQUDYEpLZbAhlYywXk2uS4yN5eFv3snP//pXADSLBb26Cle1auEQahpaLxdoaMW4XC5+/vOfs2TJEjQtMF/rKVOmMHLkSF544YWA3K870i0UjCE6GkNcPFU1Zfw8/58sr9zsd+3dX/0qJ86cYdWWLe4DQuAsUW5SKLnUerlAQyvmzTffJDo6muuuuy6gz37yySf57W9/S43qE9QquoWCATD374el1kmevYBXSj/2G4sxm0w8ef/9PPrXv+ByudAio3CcVm5SKPFlvVzgXKWVFXvP8Ktf/Yrf/e53AU+MHDNmDDNmzOBvf/tbQO/bXeg2CsaYkoLQNO5MuJY8eyFrqrb7XXvt9BlEWCys2LwZLTwcV2Ulem1tO0qruIA/6+UCtXYXP/z1Hxg0aBAzZ84MigxPPPEEzz77LBUVFUG5f1em2ygYzWzGlJbGDAaSbk7jX6Wf+s1xEULw3cU3svS9dy8cwVFa2n7CKjysO1rCieIaNIHPj3A5OL7ydRbf/VDQZBg8eDDXX389zz77bNCe0VXpcsWOTeEsLaVm8xaWmw6xpOBV/tbnQSZEDvO5ttZqZcjCG1j3yqv0jYtDmE1ETZ3a6mcrWsepkhqW7T3n9/yXa5ez/qM3WLFyNf2TIoMmx7Fjx5gyZQqnT58mLCwsaM/prHSbYsemMMTHI8xmrrSMpyCxjPSwNL9rIywWbl2wgFc+eJ9f3/t9XCXF6FYrmsXSjhIr+idFct9s/1XtV/3uY3710P0e5WJ1WnntwGuYNBPTek1jQFxgClYzMzMZMWIEH330ETfddFOb79dd6DYuEoDQNML698NYXcd3kxeSZIxrcv13Fi3m3x9/jN3hLjFwKjepQ3Hy5Em+/PJLbrzxRs8xk2Zi2YllPLvjWRZ9tIgr37mSX2/6NbuL297r5Tvf+Q5Lly69/EKFh1DOReojhMgWQhwQQuwXQjzQHs81pqWB7g4Ybq7ey//KV/tdO6hfP4YNGMCHa7PRIiJxnD7THiIqmsnLL7/MbbfdxpGqI9y76l4qa8rgfBWvjXmWTya+xK8m/R8jk0by2anP2JS/qc3PW7x4MTk5OeQ2qLxXNE0oLRgn8KCUchgwGbhPCOE7IBJADFFRGOLi0evqWFOVw3NF71Dp8l9vdNfixbz47ruIiAicpSXIhhXXipDhdDp55ZVXGHRtOrd/ejuHCvZyYNX/qN64EX3/IWKPFrKgqj9/nPoUm27dxL1j7m3zM8PDw/n617/OK6+8EoDfoHsQMgUjpTwnpfyy/v+rgINA06m2AcLcvx96TTU3x1+JTdp5v/wLb/mQbDtZxrUzZnAyP5/9x4+DlLjOn28PERVNoFutfPzaa6SMjeXFc68wzpTBf9J+ycgeYzAmJWNMSsbUoweukhKqt2yB2joANp3dxIHSA2169l133cXLL7+My+V721zRmA4RgxFC9Mc9hG2rj3N3CyFyhBA5xQEqPDSmpAAwwJzGxIhhvFOxGodsXDW7K6+cP648zP6zVXzzhht46f33ECYzjiL/bQIUwcdRXEzV2i/4+wfPoy82MsjSl6f7/5joyHivYK4hIRFps1O9YQPVJQX8fMPPWbJlCbqfJMvmMHr0aHr06MHKlSvb+qt0C0KuYIQQUcC7wI+klF7mgZTyBSlllpQyKzk5OSDP1MxmTKlp6NXV3JJwFSXOSladv5h4J5G8viUPgNe35HHnwoW8s3IltULgOHtW9YgJAVJKbCdOULN5C0V1tezcdoRZUeP4Y+8fEWnw397SEBODCLOgb/uSHw6/hz0le1h2YlmbZPnOd77DSy+91KZ7dBdCqmCEECbcyuUNKeV77flsc5/eSJuVyZHDmRw5olGV9a68ckpr3LGW0ho7JTYzU0aP5v11XyCtNjW/up2RDgd1u3dj3bePsliN11d9zg2TZrGk773EG6Mve70WHg4mM/PO92VE4gj+tONP1Dha/3d46623snLlSoqUNXtZQrmLJICXgINSyj+29/MN8fFgMCJ0yZ/7/JirYiYBF60Xq8PtY1sdLl7fksdXr7qK91evcQ9ma84AN0VA0OvqqN6yBcfZs5CczC8KXuS/lmzuvGFhi+5jiIlBLy7moYzvUFxXzNK9rd9ujo2N5Stf+QqvvfZaq+/RXQilBTMV+AYwRwixq/5zTXs9XBiNmPv08TT2tukODtadamS9XKC0xk5av+Fs2bObCocT59mz7SVmt8ZVVUX1pk3I2jqMiUm8W5HNXutxDCdg/LCWbzhq8QkMPO3itkG30juqd5tku+uuu1i6dKlyly9DyDJ5pZQbgJDOBDH1SsN26iQG4I+Fb7Kqajsjd3/TY71cwOpw8d6uYmZNmMCy7du4bcoUpMOBMJlCI3g3wFFaSu22bWhhFrS4aM7ai3m++F3iiyKZ3XNaq7JzNbMZV9V5Hoi+gfBBI9ok39SpU7HZbOzdu5dRo0a16V5dmZAHeUOJITYWzWJB2u1cFTOJGr2OXMthn2tLa+yMGzOZD9Zku7erKyvbWdrug/3sWWo3bUaLiESLjERKye8K/o2G4MQLp/jK7DmtvrcWn4DtxEnqSov535H/caryVKvuI4Rg8eLFvPvuu5df3I3p1gpGaBphffviqqpibMQgLLZYypMP+Vxrdbg47Upj6949lNdZcahevUHBWVJC3Y4dGOLiPHVfpa5KzjlKuNY5hWhXBIP69Wv1/YWmYYiKomjPNp7Z/gx/3/X3Vt9r8eLFvPdeu+5NdDq6tYIBMPbsCbqLvfnnCT+biTXuHK7w8z5bAxTXSsYMH8PyL3fgyFfb1YFGSon10CG0yChEg6bdScY4Xk9/nJLVJSycPbvNz9EiI4mrkdyadj2fnfqMw2W+rdbLMXnyZEpKSjhy5EibZeqqdHsFo0VHo0VFkWKW3Jo6CyEFmWMq+dqEvl6fmyf2ZdGcOXzwxRdIm1VtVwcYV0kJzrJytMiLbReWV27CqtsIEyY+XvsFXwmAggEwxMVzU+0Iok3RPLfzuVbdQ9M0Fi1axPvvvx8Qmboi3apdgy+EEJj79yf5wAHuGDOM6bYn6G9O9RtErBoYx6/+9kcqqquxlJdjiIpqZ4m7JlJKrIcPN/rzPGQ9xePnXqLMeZ4RJekIIRiROTAgzxMmE9HCwu1JC3j+zH/ZXbyb0cmjW3yfG2+8kUcffZSf/vSnAZGrq9HtLRhwj5hF15FSkh6W1uQORXRkJDOzJvDpjh1quzqAOIuKvayXF4s/IkaLYGHcDD7IzmbhrNkB7blriI9nkXUoE5PGt7p8YMaMGZw4cYK8vLyAydWVUAoG0CIiMCYmIuv77j5f/C5Lzr3qd/2iuXP4YMMGnKWlSIfvcbSK5uO2Xg5hiL6Ylbu/7gQba3bz9YT5RBki+DA7OyDxl4YITSMyMoG/xN7FmKTRbC/Yzr/2/4vtBdupslc16x4mk4nrr7+eDz74IKCydRWUgqnH2Ls3ep1bwdh0B8srN1Hh9P0lWzB1Glv27KHifBWuquZ9ERX+cRYVo1dWokVEeI4tLfmIWEMUN8XP4dDJk9TU1bYque5yGKKicJWX4zh7lk1nN/GHnD/w7c+/zZT/TOFvO5s3SeDGG29U29V+UAqmHlNCAtTvCl0XOw0nLlb5mTxwwU1atm0brpKS9hSzyyF1HevhQ2hRF60Xq27HKV3cnjCfSEM4H2Znc/3MWQEbqHYpWlw81v37uX/Evaz92lqen/c803tN59V9r1Jad/kuhvPmzWP37t0UFhYGRb7OjFIw9WiRkWjR0ehWK5mW3gwI68XK89v8rl80Zw7Ltm3FftZ/Q2rF5XEWl3hZLxbNzN/6PshtCVcD8OHawLtHDdHMZqTThe3kSRLDE5nWaxoPTXiIgfEDKam7/AvEYrEwf/58Pvroo6DJ2FlRCqYB5j590KvdLs+86InsrjtKgcP3G2zu5Mls2LWLuvIydKvvoWCKy2PPPYWIuBjYPWU7R6GjDABNaJzMz+dccTFTRrt3eGR90y9XRQV6TQ26zYZ0On3euyUY4uOxHz+OXt+xMCM2g7eue4vBCYObdb1yk3yjFEwDjElJUJ87d1XMJL4a5z8lPTE2lmEZA9i4dy+uStXlrjXodjvO4uJGO0fPFL7O93J/55m8+dHabK6dMQNNCJxlZbhKijEkJGDq0xtDbAxCgF5Tg7O4CGdxEa6qKqTe8h0hYTCALnGcyW90vNJWydHyo5e9fsGCBWzatEkNZ7uEbp8H0xAtOhpRX5vUy5zMQz1va3L9VVOmsHLnTuYXF2HqkdJOUnYdnKWloEvP1vMZexE7ag/xg+SbPP15Ptu4kfuuvx5XaSnmfn0xp6f7zD3S6+pwVlTgPHcOR2EhuFwYkpJbtK1tiI3Fduwo5r59EEb3P417Vt6DUzr573X/bfJeUVFRzJw5k+XLl3Prrbe25I+hS6MsmAYIITD16Y2r2t0EXJc6u2qPUlRvsl/K/KlTWZGTg6OgQJUNtALHmTOIBrGXnNqDAEyLcrtDFUVF7DxwkLnXXEv0nNmEjxzpN7FRCw/HnJpKxLhxxFx5Jaa+fXGVtiwAL8xmpN3uVlD13DjoRg6VHSKn8PID/6655hqWL1/eomd2dZSCuQRTSgrUN3QudVZyb95TfFS53ufakQMHUmO1cuzECfQaNbu6Jeh2O87CwkbB3ZyagyQb4+irJeEsLuKL3buYPHkSSRMnNFp3OYTRSPiwYRji43FVVrRILi06BtvRox4367qM64gLi+O1A5dvLjV//nw+//xz9Fa4aF0VpWAuwRATgzAYkC4XyaZ4xkYMYuX5bT4tFCEEV18xhRU5OS3+Ind3nGVuq1DUbz1LKfmy9jDjwgail5djGTKE7FOnWHD99a26vzAaiRg3DjStRTVjmsWCXlWNq37InsVo4aZBN7H29FpOnz/d5LXp6enEx8ezc+fOVsncFVEK5hKEwYCpl7shOMCV0RPJtRdwxOY7FfyqKVNY+eVOnCoHokU4zuQjwi9aJUII3kh/gruMswnPGo95wAA++/xzFixY0OpnaBYLkRMmoNfWtmielYiIwHb0YmD31iG3YtSMbDy78bLXLliwQLlJDVAKxgem1FSk3QbAnJgsDBhY4ScnZvbECWw7eIDK3DykmpXTLKTdjrOwwMvtiXWZSYtKw5SSwv79+zEYDAwZMqRNzzLExhI+biyusrJm7y4ZoqJwlpbiqt8RSo5I5vMbP+eWIbdc9toFCxbw2WeftUXkLoVSMD4wxMWBpiF1nVhDFJMih7GtZr/PtTGRUYwfNoy1O7/0WD2KpnGUlYGUHvcI4OWSj1lWlE3Y4EEIg4Hly5ezYMGCgBQ3mtPSMPXrh96CoXnCHIbt5EnPz8kR7pE5NpetyetmzJjBnj17KFeN4QGlYHwiTCaMKSno9cWPj/b8Ji/3/4Xf9VdfMYWV27d74gqKpnHk5yMsF2cZOaWLN8o+Y59+GnPPngAeBRMowtL7I+22Zu/2aTExOM6caVRrtnTvUhZ+sBCH7r/A1WKxMH36dDWYrR6lYPxgSktDWt0jR5NN8ZiE/5Shq6dOZeWXX2JX7Rsui7ywe9Qgue6QNZca3crkftMRJhNVVVVs376dOXNa33v3UgyxsRiTk5sd8BWa5rZijh/3HBsYN5D86nxWnmpaeag4zEVCPXjtZSFEkRBiXyjl8IUxLs5T/AiwrHIjPz79Z59vwEH9+mE0Gtm7a5dq33AZnOUVoOuN3KPt1W73c/JQd+3RmjVrmDRpElEBbuYVlpnZsh2l2Fgcp097rJjpvafTP6Y//zrwryYtofnz5/PZZ5+p3ChCb8G8CswPsQw+0SIi0CIj0W1un9umO9hcs5dce4HXWiGEezdpew6uFvj53RFHfj4izNLo2I7z+8iM6EdSTA8g8O7RBQyJiRhiotHr6pq1XgjhtmLqd5Q0ofGNYd/gQOkBdhTu8HtdZmYmUVFR7N69OyByd2ZCqmCklOuADhu4MKaled54U6JGArCx2veX5uopU/g8J8eTP6HwRq+rw3E2H62BZaLX77xN6zMDcOfDBEvBCCEIGzQIV1XzXwJabCz2/HzPi+P6Adc3K/FOuUluQm3BdGhMSUmerN6epkQyw3qzsWaPz7XTx41nz4njlDbw2RWNceTng9AauUd6ZSX/GPkEP5n0MAAHD7rLBYYOHRoUGUwpKWhhlmbnxQgh0MxhWI+4rZhwYzhPTX+KRyc92uR1F9yk7k6HVzBCiLuFEDlCiJzidp5FZIiJAYEnf2Jq1Ch21x6jyuVdFhBhsZA1fDjrt2zxuFWKi0inE9vx4xhiYxsd1512zH36eLajA7k97QthNBKWmYnrfPMH52mxsTjPnfMM25vSawo9I3s2ec2sWbP48ssvqezmA/o6vIKRUr4gpcySUmYlJye367OFyYQxMQlZ77PPiBrLrOhxVOu+645mT5hA9s6dLcq36C44Cgu9xu1Kp5MHql7i94cujg0JlnvUEFPvXiBEsxMjhRCIsDCsR495jh0qO8T3V32fcqvvfJeIiAimTp3KqlWrAiJzZ6XDK5hQY0xL9fTqHR6ewZJe95BqSvK5ds7ESXyxezcO1UazEVJKbEePNmqLCVBVVcoexymizTEA1NXVsWXLFmYHsXsduDvYmTMyPJm6zbomJgZnfj623FyklBiEgfX56/ng2Ad+r1FuUui3qf8DbAYGCyHOCCG+E0p5fGGMj/c0obpAvr3Y0xCpIaMHDaKwooLT+zrcrntIcZWV4TpfhRYe3uj4l7UHcaEzKXUSABs3bmT06NHExMQEXaawPn1AdzW7fEAIgSEpibrdu7HuP0BmTAZZPbJ4+/DbfkeezJs3j9WrVwdS7E5HqHeRbpVSpkopTVLK3lLKl0Ipjy+0KPcY0wttGddV7eTGEz/joPWk11qDwcCM8eNZs3lzs7dCuwO2Eye86o6klKyy7SHWHMu4lHEArFq1innz5rWLTFpkJKbefVpWPmA0YkxOwZ57itrt2/laxmLyq/PZmO+7CHL48OHU1tZy4sSJQInd6VAu0mUQQmBK7enZrh4dMRANwQY/29VzJk5k7a5duM6rcSYArupqnAUFjbamAWqqy9ngOMjV/a/GZHDHZVatWsXcuXPbTbawjHT0FpQPgDvD15iUjLOigol5ZpLCEnn78Nu+1wrB3Llzu7UVoxRMMzCm9PBUV8caohgZnsmmat/b1bMnTHTHYYpU+wYAe14emExeu0K6tY7vZdzOjYNuBKCsrIwjR44wefLkdpPNEBODqWdP9FbMtjLGJ2DWzNwRNpPxCaP8Kqnu7iZdVsEIIe4XQsS3hzAdFUNcLCA8X6KpUaM4YjtNkcN7ByG9Vy/CwsLYu9V3k6ruhG63Yz91CkNMrNe5CGHmW2O+y7BE9zC17Oxspk6ditlsblcZwzIzPTVnLUWLjOTGqOl8tTwT6Sc14YIF01273DXHgukBbBdC/FcIMV8EK0GhA6OFhWGIjfF8iaZGjQJgk4+kOyEEsydOJHvbVs8o2u6K48wZd1sGg6HR8dK6Elbo+7GaL36V2jP+0hBDXByG+IQW1Sg1uj4mhjp7Le9n/wNrnbcl1LdvX+Lj49m7d29bRe2UXFbBSCl/AQwEXgLuBI4KIX4rhBgQZNk6FKYGZQMZ5l48mXYPs6PH+1w7e4I7DuPsxvkw0uHAduQIhtg4r3Ofl27kibLXKKi9WNe1evXqkCgYIQSWwYPQa1rfy+eAuYRfF73CJ+tf8jmjae7cud02H6ZZMRjptvUL6j9OIB54RwjxdBBl61AYEhKQujsxSwjBvJgJxBp8V/vOzMpi8/79WC+ZsdOdsJ87h3Q6GyXWXWBl3ZcMjslkQJz7HZWbm0t5eTkjR45sbzEBdxGkFt38IshLyYoYSj9zT94pXkndfu/GZN05DtOcGMwDQogdwNPARmCklPJeYDxwY5Dl6zC4m4EbPXkTVa5a3ipbyTHrGa+1SXFxZPTuw+b167tlHEa6XNiOHEHzEXvJtZ7loOsM12VebOa9evVq5s6dG7TZ05dDaBphgwaht6AIstH1QrA4bjb7nXnsPbYJR1FRo/OzZ89mw4YN2FvQF7ir0Jy/0QRgsZTyainl/6SUDgAppQ5cF1TpOhDCYHA3LKqPq0gkfyl6m+wq32X7sydOZM327a327TszjqIiZF0dWliY17nPSjcgECzIuNZzLFTuUUNMPXq423O00oq5JnYKYcLE5+IAdXv3NeoLlJCQwKBBg9iyZUugxO00NCcG8yspZa6fcwcDL1LHxZTa07PjEGOIZKilP9trD/hcO3vCBL7YvatF6ehdAanr2A4fRov2nY171JpHVsJoekS6e79IKT0WTCgRBgNhw4a12oqJNkSQFTGU03oJ0mbFeklVfXd1k1QeTAu4NGA5IXIo++tOUOPyfutdMXo0+06e7HbtG5zFJehV3mUB4FY+v4u+gz/P+pPn2P79+4mMjCQ9Pb09xfSJKSUFLTa21Vbnkl738Oc+P8YQn4D96NFGL5fuGuhVCqYFaJERiAa9RCZEDMOFzs66I15rwy+0b1i7tlXD2DsjnqLGSN/Bb9f585h79yIm+mKxaKi2p30hNI3wYcM8o4NbikUL89xHRERSu3evp2J72rRp7N69m/PdbGdRKZgWIIRwD2Wrj8OMDM8kQgsjz0cbTYCZWRP4YufObhOHcZWX4yovb9TQ+wJSSr5Z/AxvOTc3Ot7e5QGXw5CYiCk5qdVK5u9F7/BY/j8wREXhqqhwZzID4eHhTJw4kXXr1gVS3A6PUjAtxJSUhHS6A3hhmolPM//M1xOu9rl2ZlYW6/bs8TQq6urYTpxEWCw+zx2pPMpxVwExsSmeYw6Hg/Xr1wd0ekBbEUIQNmQIsramVTuAOpJ1VbuocdVhTEjEevCgJ3DcHeMwSsG0EENsLEjp+fJZNP+p7eOGDiWvsIjCI94uVFdDr6nBWXAOLTra5/n1lTsQCGb1meU5tn37dtLT00lK8t1fJ1QY4+Mxpqa2qkZpatQonLjYVnsAYTQCwjPArTvGYZSCaSHCbHYn3VmtAJQ7q7gv7xlWnd/utdZkNDJ59CjWrlnT5eMwtjNnQDP4bHUp7XY2OA8zMmkkSeEXlcmaNWs6lHvUEMvAgUhrnc/M3KYYFT6AKC2cTdXu0gBDfDz2EyfRa2oYP348p0+fpuiSPJmujFIwrcCUmupJLY81RHLUeprNNb5rTWZmZbFu164uHYeRdjv2Eye9+u1eoKDiNIccp5ndt3Gnuuzs7KB3r2sththYLKNG4SopaZGSMQojkyKHs7lmD/LCeFyjAdvx4xiNRqZPn87atWuDJ3iAePvQ21Ta2u7aKwXTCowJCZ6hbJrQmBA5lO01B3z67DPHZ7Fu926cXTgfxlFYBC5nvUvQGKnrCAl3DvkG8/pe3C2yWq1s3bqVGTNmtKeoLSKsXz8so0biLClukZKZH3MFV8dMxibdu42G2DjsuXm4qquZM2cOa9asCZbIAeGTE5/w5NYneffou22+l1IwrUCLjkYYjZ4tyAmRwyhylpNn9+4BM3LgQIoqKzm9z7tGpSsgdR3bsaM+ywIAXJWVpKWP4MFJj9A/tr/n+JYtWxg+fHi7tMdsC2Hp6YSPHIWzpKTZTcKnR4/h/pSvNd62NpuxHT3K7Nmzyc7ODqbIbSL3fC6/2fwbxqWM445hd7T5fkrBtAKhaRh7XOxyNyHC3dNkW623EjEYDEwfO44vsrO7ZBzGVV6Oq6raZ1kAQJ29hp3hxV4D49esWdOhdo+aIiwjnfCRI1qkZJzSycG6U56ftdhY7GfOMLxfP0pKSsjP75iFsNl52ZgMJp6a8RRGzf889uaiFEwrMfa82OWulzmZq2ImkWSM87l2Zn3ZgN7K3IqOjK9+uxfQa2rIsZzlexsfYGfhzkbnOpOCAQjLyCB86BBcJcXN2r5+rfQzvp37JOVO906UEAItzIL9+HFmzZrVYa2YO0fcyQcLP7js3KfmohRMKzFeEtB8Iu1uv/1hZmaNd8dhulg+jL9+uxfQa2vYKE4QY45hbI+xnuM1NTXs2rWLqVOntpeoAcGcmYk5PR1XyeUHAE6OHIFEsqVB8F+LicF57hwzJ0/ucHGYzWc3s7/EbYE33OlrK0rBtBItIgItIqLRFMcqVy2VLm8rZUj/dOrsdo7v3NWOEgYf+5kzYDT63pp2uXBqgg2l25neezom7WJfmA0bNjBu3Dgi/Fg+HRUhBJZhwzD26IGrrOkZ5IMtfUkwxLChQe9mIQTCEs7U/ukdSsGcrDzJQ188xG+3/jbg7UVCPRdpvhDisBDimBDiZ6GUpTWY0np54jDVrlrmH/0R75d/4bVOCMGM8Vms7UJxGOly4cjN9dlvF9x1R4fj66iwVTC7j/f2dGdyjxoiDAYiRo9G1JcC+EMTGjOjx7Ghelejl44WHc2AMDM2q5WTJ71H37Q3JXUl3LvqXoyakd/P+H2jl4X9zJk2f19DpmCEEAbg78ACYBhwqxBiWKjkaQ3G5CSoD/pFGSLICEvz275hZlZWl4rDuCorkXaHz61pAJwOtsrjGDUjU9Mau0Jr1qzpsPkvzUGYzUSOHw8Grcn8phvjZmOTjkZJmEIIhMnMjHHjQm7F1DpquX/1/ZTWlfL3uX+nT3QfzzlnWRnWAwca9bVpDaG0YCYCx6SUJ6SUduAtYGEI5WkxhpgY0IRHy2dFDGVv3TGsunfnsplZWazfvRtHF8mHcZw9i/AzAUC32dAiI/nBpJ/w9nVvE2W+GKOprKzk4MGD7TqeJBhoERFEZmWh19T43VnKtPTmxX6PsihuZqPjhrg4pg3IZM3Kle0hql/+c+g/HCg7wNMznmZE0gjPcanr1O3fH5DhgaFUML2A0w1+PlN/rBFCiLuFEDlCiJzi4ssH19oT96S/i13uJkQOwy6d7Kk75rU2vVcvjEYjB7d5lxR0NqTTiePMGb91R3rVecwZGZgMJgbFD2p0bt26dUyePJkwP9vanQlDXByWoUNwlfqPx4wMz0QTjf+ZCU1j+tgxrFmzJqQtVe8cfidLr1rqlWHtOHsWvaLCv3XaAjp8kFdK+YKUMktKmZWcnBxqcbwwpaYi6zX9mIiBGDCQU+Pd6E8IwYysCWSvzW5xfUtHw1VRgXS6vMaRgPvth4SnT7/Ci3te9Drf2d2jSzGnp2OIj8PVRJ+X10s/4xf5/2x0bMDgIRiBgyEYZ/Lx8Y8pqi3CoBmY0HNCo3PSbsd64CCaj2kQrSGUCiYf6NPg5971xzoVhrh4wP0WitAsPJ72Xa6Lm+Zz7awJ7rqkzt5G05Gf7989qqmhIsnCeyc+pMJW4XW+s+W/XA5hMBA+ZgzSZvP74rBLB6uqtnPKds5zTDMamT5qFKs//LC9RAXg81Of89iGx1i6d6nP8/bcXKTD7jdxsqWEUsFsBwYKIdKFEGbgFuCjEMrTKrTICIQl3NPlbl7MBPqae/hcO2P8eNbv3Yvt9Gmf5zsD0unEftZ/WwZpreNTuRundHLToJsanSspKeHUqVNkZWW1h6jthiEqCsuI4X63rr8SNxOTMPJOeeOg7qzJk1m9YkWbA6nNZWfRTh5b/xhjU8byYNaDXuf12lqshw9jiE8I2DNDpmCklE7gB8DnwEHgv1LKTlewc6HLnat+N8GuO/i8cguHrKe81vZK6UFCbBw7N23utG6Ss6zMXdjoY8SIdDrRTUbeO7OcSamTGtUeAaxdu5Zp06ZhDIBv39Ew9+2LMTnFp3WaYIxhbnQWn57f2Kh/88yJk9iwZw/Ws2eDLt+pylPcv+Z+UqNS+evsvxJm8LZQrEeOuPOafLi+rSWkMRgp5adSykFSygFSyiWhlKUtmJKTob7LnRCC3xX8i08qNvpcOzMri3U7v8RV7j3XujPgOJOPCPPdtc5VWcn22FIKagu4efDNXue7mnvUEKFphI8cgXTYfe4q3RQ/l1rdxqfnN3mO9e7Rg9joGL5cvjzoL5xnc57FIAw8P/d54ixxXucdJSU48vLqXf7A0eGDvJ0BQ2wsCPd2tUkYGRMxiO21vie6zMzKYv2+fTja4a0VaKTdjqPgnN/SAFwuevcayk2DbmrUue4CHWE8STDRIiMJGzgIV4X3y2N4eAbfSryOkeGZjY7PmjiBL3J2YA9y8eOS6Ut44coX6BPTx+ucbrdTt2sXWkysz6zstqAUTAAQJhPGxETPbtKEiKHk2s9R5PD+ok0fN47N+/dTe/pMp3OTnOUVoOs+3SO9rg5DXCzD+ozj/674v0alAQBnzpyhtLSUUaNGtZO0ocGc3h80g8+4yveSFzHE0q/R1vTM8VmsP3gA2+HDAY/FuHQX/97/b2wuGzHmGAYnDPZaI6XEeuAgOBw+R820FaVgAoQxLQ297mI+DECODysmOT6ePj17suvwoU7nJtnPnEZYfH8J9ZpqNkUWcLT8qM/zF7anQzUetr3QzGYsQwb73SmsdtXyozN/YnV9du+M8ePZtHs3DqsV+2nvMcStRUrJ09uf5pmcZ8g+7b9y21FQgD0vDy2Agd2GdO2/7XbEGB9/YbeazLDexBmiOGLN87l2ZlYWX+zb36ncJOlw4Cws9D2SRNdx6A6ePPI3/rbzbz6v78rxl0sx9+6NCLc0KoT1nBMmanUrvzn3MsesZ0iOj6dvaiq7z53DevgweoDmV7924DXePPQmdwy7g/n95/tco9fVUbd7N4b4+IC7RhdQCiZAaFFRCLMZ6XSiCY23M5bwox63+Fw7c3wW6/ftxX72bKdxk1zV1aBL3+5RdTUbI89Rbivna4O/5nVeStmhG3wHGmE0Yhk2DL2ywuucWTPx27R7idTC+Vn+3znvqmFW1gS+2LkLpI4jz/dLqSWsOLWCP+T8gSv7XelzOxrqywH27QMh0PzkNAUCpWAChBACU1qqp5gx1uAnEApMHTuW7fv3Y62r6zRukqu0FPy4N9Jm5aPaLaRFpjElbYrX+WPHjqHrOgMHDgy2mB0GU48efsfQJpvi+V2v71PgKOX/zr7A9KxxfLEjB0NsHNajx3xaPs3F6rTy+22/Z1TyKH477bdeZQpwYQLnMZwFBRgDvGt0KUrBBBBjSgrS4TZxrbqdX+T/k08rN3mti42KYmh6BjknTgZ99yBQOAoLfXauk04n+aKSbaVfsmjgIp9f6AvWS7DM8I6IZwxtTbXPeqNREZk82OPrnLDlM2RkBjsOHMDqdILUsZ/KbfVzLUYLL171Iv+Y9w8sRu90AqnrWPcfwHb4EIak4JfeKAUTQAxxcYD7DWHRzByy5rKmKsfn2plZ41l/YD+Oc+faLZOztUi7HVd5uc+pja7z5zkV7yTaFM2izEU+r1+9enW3ib80xJCYiCk1zee2NbgzfN9If4IBsb0ZPmAAW/ftwxAbh+34MfT6uVvNZcu5Lfxz9z+RUjIgbgAxZu9m6tLppG73bmwnTmBITvHp7gYapWACiGY2Y4iP9wxlmxw5gh01h7Dr3gpk5vgs1u3YAU4njrKy9ha1RbiqqkAI3xaIy8nVw7/C2pvX0iPSu0RC1/VO3WCqLQghCB8xHKFpPt0eIQTRhghcUifp5mT+k/s5wmhEALZTp5r9nO0F27l/9f2syF1BndN3iwVpt1O7YweOs2cxprSPcgGlYAKOKS3NM5RtUuRw6qTNZ/uGyaNHs+foUWqEhiO39SZxe+AsKQXNO31ct9moizAioqMxG3wHCvfu3Ut8fDx9+ngneHUHNIuF8DFj0MvLfXaHk0i2nCxBSzWwe/AJdtQcQouLx378eLP6sWzI38B9q++jV1QvXrzyRSJM3m6sXltL9datuMrKMCYlt6urqhRMgGk4lG185BCMGNhcs89rXYTFwtghQ9ly9AjO4mJPT5mOiLOwwOf2tF51nl/VvMUP1vzA77XdaXvaH6aUFMyZA3wWQ+7KK+dvq47zXfN3sBfZefjM3zhkPw1C88y09seyE8u4f/X99I3uy9Krl5IYnui1xlVZSfWmTcjaOgwJ3ueDjVIwAcYzlM3pJEKzcG3sVJL9jTPJGu92k4TAUeg9tK0joNtsuCrP+yzfL3BVsLnsS4YmDvV7fVcvD2gulkGD0CIjG+0qSSSvb3FvSy/bWUmPFdEYnBrfzn2SV/UNnpnW/hAIxvcYzyvzX/E5CcBRVETNxo0IzeCJD7Y3SsEEGKFpmHqmeiySR1O/yS0JV/pcO3N8Ftnbt6NFRWM7cTKk3c38oZ8/D5q3Sa3b7Xzq2gXA4oGLfV7rcDhYv349s2bNCqKEnQNhMhExdqy7xWZ97tOuvHJKa9y7jqU1dob0HMHItX25LeFqRoQPAKOBkqP72F+ynyPlRzhUdoj9pfvZXbwbgGsyruGFq14g2ty4dYaUEltuLjVbtiKion1an+1F16ub7wAYU3tiP30a6seiOqWT865aEoyNI/sTRozgVH4+JTU1JDiduCoq3BnBHQhHSQkYvL8mjqrzLLPmMCVtCr2ivDqdArBjxw7S09PpiJ0IQ4EhNpawwYOwHT2GISmR17fkYXW4K6+tDhfFIpWjm/7HH+97GHBvKb965HVe27220X2MmpF3rn+HAXEDvNICdJsN67592PPPYkxMDEjby7agFEwQMMTGNnrrf+Pk42SEpbGk172N1pmMRqaNHcfanO3cOHESjjNnOpyCcRb4jr9ssx6k0F7CTwc95vfa7ro93RTmPn2wHTnKrlOlHuvlAlpUT3LPFVBUVkZKQgJC01gYMYUhKSOw9E/HIAxoQiM2LJaM2AyvezsKi6jbvQvpcrl3ijpA3pFykYKAZrFgiI72bE2OCB/AtpoDOKV3n5A5kyaRvW0bWnS0ew5NB8qJ0a1W9Npar1RyqeuMDsvg91N/y6zes/xev2rVKhV/uQQtPBxT7168t+6wx3q5gN0FiT0HsG7HxdypXonpzKjsyfTa3sxOvoLZfWczrse4RspDr62lbt8+arduQVjCMSYkdgjlAkrBBA1TWpqnbGBy5HCq9FoO1HnvCsyZOJE127a50/BdLrdL0kFwVZ737Ig1RK+tJSalN9dmXo/JYPJxJVRXV5OTk6PiLz7Y6oikqtr3rmF86iD+t2a952ehaRhTemA/lUtVdjZ1Bw6g19ai22zY8/Op3riRqjVr3M2iklMC1ks3UCgXKUgYEhNBHgbc7Rs0BFtq9jEq4mLDIYmk3BWFEILDp04xKCUFx6lTmFNTQyV2I5wlxQiTd37Ltso9nLLAN1yjfLZeBPjiiy+YMGECkSEMMHZEpJQ8+cUZog3hWJw2rMbGf35xPQey9vN/okvdE18RmubuN6Tr2HNzsZ844V4sBFp4BIZ2zm1pCcqCCRLuoWwGpK4TY4hkePgAtlySD7Mrr5w/rTrC6GGjWbNtGyIiAmdpKa5q/1uT7YWUEse5Ap/1R+9aN/Gf/I+8mko15PPPP+eqq64KpoidknVHSzhRUkt+dA+iXDY0QaNPdGwKOvDxVu/cKaFpGBMSMSQmYUhKxpiUjBYZ2WGVCygLJmgIgwFTjxSc5RUYoqL4XtJXMIqL2bANcyCsEX1Zs20r37/5ZoSm4Sw4hyEz09+t2wVZV4e0WRGXTA+orCtni+MIXx94m8/CxgusWLGCN998M9hidjr6JUTw0NWDQXeRtEfHZQ5DGhsr6uojWRw4uIuFk313/2uvNP9AoBRMEDH27Inj3DmIiiIrsnEyWsMciIjEAaxe8wZ2hwNTdAy2U7mYMzJC+kVyVlb6jL9kl2/FiYsF6Qv8Xpubm0tZWRljxowJooSdk/5Jkdw32/3ysKWbsR48gPHSquayOfz744+AO9pfwAATkm+wEOImIcR+IYQuhOhaQ3IaYIyNbfSPdFftEZZXbvZYLxd2EaQxnIjoZLbu24swm9GtdbgqK0MlNgDOwkKf0wNW1u2kT0QawxOH+7125cqVXHnllV2+PWZbMfVKA/CqUZqZlcXm3buxtqEvTEchVN+AfcBiYF2Int8uaJGRiPAIz1C2Fee38vuCf7Eh94xXDkRi2mDeWun+49BMZhxnAteftaVIXXfnv1wSf3G4HBjRuG7A9U36/StWrFDxl2aghYVh7tfP62USHxPD0IwMNu/ZEyLJAkdIFIyU8qCU9VssXRxTWqpnKNviuNnYpIO/H1/plQMR33Mgyze5rRstJsadExOg/qwtRa+uds+eviQL1GC185cBj3Dv2Pv8XutyuVi1ahVXXum7PELRGHOfPp6ZWg2ZO2kya7ZuDYFEgaXD27BCiLuFEDlCiJzi4uJQi9NiTMnJ4HLXnmRaejNASyc/YS+SxvGN+B4ZlJWeY/2BPHfsxeXCUep7FGmwcZaXgw8LpbymGGNaWpPWS05ODr169SItLS2YInYZDLGxGOLivKrp506axKqtW0IkVeAImoIRQqwSQuzz8VnYkvtIKV+QUmZJKbM6Y02LISYGpERKiURiODUIh+U81vjGrTINBiPxKen86b3VbismIhJHC5oOBRLnuXNeM3LO2ou5oWwJn1dta/Ja5R61HHNGhicp8wJZw4ZxuqCAwhC9ZAJF0BSMlHKelHKEj8+HwXpmR0SYzRgSEpBWK3vOVGLPTcNUG4turvXKgUjpNZgjR/ey50ylJyemqXL9YCAdDpylpYhLFMyKii240BnXZ1KT169YsYKrr746mCJ2OUzJyWA0Nho5azQamTF+PGu2dW43SW1TtwOm1FSsBw/SIzqSWyakI/kJop+Afo3XnU6bxR+f/z09osPcbojQcBQWEpbhXdgWLFznz4P0zrXIPp/DiMiB9I7u7ffayspKdu3axfTp04MtZpdCmEyY+/XFnpvnblhWz9xJk1i9dSu3LrgmhNK1jZAoGCHEIuBvQDKwTAixS0rZZV977qFskp6x4XxljLu1gS51ChxlpJkvNgqSo9P46z91qiqL6RnbFy0mBtvJU5jT09stW9NZUgqGxu0xixxlHHae4YcZdzV5bXZ2NldccQXhQRhB2tUx9+6N/fiJRsfmTprMkhdfRNf1TrvlH6pdpPellL2llGFSyh5dWbkAaDExCM3QyAR+quA1vpv7W5zy4uA1IQTzp05l+Xp3sZtmNiPrav2OIQ0GjnPnvNozrK92NziaPaDp2IqKv7QeQ0wMhvj4RsHe/mlpxERGsu+Yd0/nzkLnVIudDKFpGHukNPryzIgeQ6mrki+qdjZau2DqNJZv3HDxWpMZRzvNTtKtVvSaaq+K3KnmoTya8HUGJA/xe62UUtUftRHzgAxPw/gLzKl3kzorSsG0E8aePZG2i7NuJkeOJNWUxLsVjQeTz5owgZ2HDlFRVQW4rR97bm67NAV3+SkPSLKZWTzoxibdtEOHDmG32xk5cmQwRezSmJKSwGBsNE74yslX8Pkm7+F9nQWlYNqJS8sGDEJjUdxMvqw9zEnbWc/xCIuFqWPGsGqLOwdCaJq7w3w7bFk7CwsR5sbWy7664yyv3YYeH+3nKjcfffQRN9xwQ4eu7O3oCJMJc3p/XOcvZvbOyspiz5HDlLVz6cgFJdfWv0+lYNqJC2UDeoPs3Otjp2MWRj6p3Nho7fyp0/isgZtkiI93d5gPohUjpcRRUOgVf3mvfC1/rfkEU0xsk9dfUDCKtmHu1QvpdHoawIdbLMwYP54V7WjFSClxlpZiGTYMYfY976q5KAXTjpjSUhvltcQbo3m+70+5N7nxyNX5U6eyYvNmXPVBYaFpYNCwnWh6Tk5b0KurkXZ7o/IAl9TZVL2HKVEjMVm8+8JcoLi4mP3796vudQHAEB2NKTXNPc2hnmumT2fZhvVNXBVYXKUlhKX3x9y/f5vvpRRMO9KwbOACw8MzMIrG2QJ9evYkLTmZbfsuNh0yxMVjP9X0nJy24KyogEus4X11x6nQq5nVa1aT1y5btox58+YR1sHaNXZWLAMzkTarx4q5espU1mzdiq0datOcFeUY4uOxDB0aEHdXKZh2xFM2cEl5/qeVm7g/79lGc5EWTJveyE0SmgZG42Wn/bUWR24eWmRUo2Prq3dhxMC0jNlNXqvco8BiiI3F2MCK6ZGYyJD0dDbs/DKoz9VrahBGIxHjxgVs3IlSMO2IMJsxpqR4xVIEsL32ADm1Bz3HFkybyvKNjWMzhtg4945SgK0YvaYGV0W5V/1Rnq2AsaYMYhP89wi2Wq2sXr2aa67pvNmmHRG3FWPzvHSunT6DT9YFr7uJbrej19YSmZWFZvHuA9RalIJpZ0y9eyMvGWo+J3oCcYYo3ilf4zk2fugwisrKyD17cYfJHYsxYDvROOOzrTiKinxWT/8u/ls8k/kQwuA9+P4Ca9asYfTo0SQleY8uVbQetxWT6rFirpk+nU/Xrw/K9E/pcqGXlxE+fpx7plcAUQqmnTEmJgKy0RclTDOxMG4G66t3UeBwV88aDAauvmIKn23ybcW4GgQB24KUEntuLlp0jPc5ax3RqX2bvF65R8GjoRUzuH9/LGYze44cCegzpK7jLCnGMmJEUKZZKAXTzmhhYRiTk5GXuEmL4mYB8H7FWs+x+dOm8tmGxgpGaBoizIJ1/36vWE5r0Kur0atrvLJ3HzrzV56vXo4hPs7vtVJKPv74Y6VggkRDK0YIwTXTZ7BsfWB3k1wlJYRlDMCcnh7Q+15AKZgQYOrTB72usYLpaUrknuTFTIoc4Tk2d9IkNu/ZTfUlysgQE4OzpARHA/eptTgLCryG2x+znmFT9R6MRnOTg9O//PJLoqOjGTRoUJvlUPjGMjAT3WpF6nq9mxS4OIyrtARjzx5Yhg4JWoKkUjAhwJiQABIvf/qOxGsYFzHY83NMZBQTho/wZPU2RIuLp27ffs942tYgpcSWm9fIPdKlzlOFrxEjIrgt46tNfvGUexR8DLGxhGUOwFVcxKSBA8krKCC/qLBN95RS4iwrRURFETF6dJMxtraiFEwI0CwWjEmJXm4SQJGjnOeL3/NUWS+eO5d3V6/yvofZjJA6tjb45HplJdJa12j29MeVG9hbd4zvRywguffAJq//6KOPuP7661v9fEXzCB82jMipUzEaNK4cN45lX3zR6ntJux1nURGmlBQiJ05sc6bu5VAKJkT4cpMAjtlO86/SZXxauRmAhbNns3rLFqp8bE1rcfHYT51y99BtBfaCAjBczHdwSicvl3zMmPBBLIiYgCE+3u+1eXl5nD59miuuuKJVz1a0DGNiIlHTp7Nw0SKWZWfjLC9vcQzOWVGOq7qaiPHjCB83LqDb0f5QCiZEuN0k6eUmXRE5kqGW/rxa+glO6SQhNpYpY8awzEcOhNA0tMgo6vbsadRrpjlIXXcPTI+56B4ZhZF/9vspj0bfhDk1rclkq7feeotFixZhDFBCluLyCKOR6+68k+3HjlEdFYmrvAxnSbFfN1m6XLiqq3GWlOAsLsKYkED0zBmYe/Vqt6JUpWBChBYe7u7Ve0lOjBCC7yTdwFlHCZ9VumMvN115Fe+sWun7PpGR6OersOfm+X1WSV0Je4v3sqNwB1anu2VERfEZam01HiVS7qxCSkmqKYlerihMl5kK8MYbb3D77bc3+/dVBIbo6Gjmz5/PJ/v3Ez1vHpYRI9x9lEuKG3xKcJaUoJ+vxBAbi2XkCKKmTSMiK8vnrPFgol4/IcTcpy91e3Z7/aVPjRzFEEs/Xi1dxvzYK7hm+nR+/MzTlFZWkugjEcqQkID14AFMPXs0uteR8iP8a/+/+PTEp56YzieLPqFfTD/+u+d1ni97m4G1fRgVnsnWmv2MCh/IYz3uACEwNrE9vWfPHsrLy1Xv3RBx++2389RTT/G9732PsH79MPfp0+hFJaU7b1KEh4d8jrWyYEKIMSkR8N5NEkJwV9JCRoVnUqtbiY6MZO7kyXyUne3rNm4rxGik7sBBz712Fe3ixo9uZGXuSm4ecjPPzXmOF658geTwZFzVNYypSeQb8fOJ0Cx8WLGeXHsBs6LHotfWYuzRo8ng3xtvvMFtt93WafvEdnbmz5/P4cOHOVGf0e12lSM9H0OU+7+hVi4AIhipx8EiKytL5uTkhFqMgFK9cSPSZm8y3wTgw+xs/t87/+PTv//D53mH7mRV/hqc/dO4Jevb6FLnP4f+w3UZ1xEb1tjqqcnZgau0FENcHOAO7pY5z5NiSsBZVER41ni/WZ26rtOvXz+WL1/OiBEjfK5RBJ8f/OAH9OzZk1/84hehFgUAIcQOKaXXnPnQq7hujjkzE73Wf/HiIWsub5etZNbkLPYcOUJBSUmj8zWuOt4uW8VNJx7j8Zq3+ejo++g2G5rQuG3obV7KxVlWhvNsPloDV8sojKSYEtzWj6DR6IxLWbduHYmJiUq5hJjbb7+d119/PSi1SYEkJApGCPGMEOKQEGKPEOJ9IURcKOToCJiSk92d7nzsBEgpea30U/5U9Ba35z/OpG+O5X+rV3jOf1q5ieuPP8Sfiv5DD1MCT/e6n79H3+23pYPUder27UOLiva5iyBrazEmJXuVDTTk9ddfV8HdDsCkSZNwuVx0dIs+VBbMSmCElHIUcAR4NERyhByhaVgGDWrUwcxzTgiW9LqXf/R9hJ6mBPKzyniz32qOWN07Rn3MKcyIGsPSfj/n//X7GTOix2BKTMZ+9KjPUSeO/LPolZV+3TG9rhZTH/+D1axWK++99x633npr635ZRcAQQnismI5MqOYirZDSMxBoC+D/W90NMPXsgTAaGnWTb8i4iMG80PdRfp/6fezVdt45627rMDI8k1+nfZcR4RcnP17IjaneuBHr4SMey0i327EePIAW5zt5TkoJUjbpHi1btoyxY8fSq1ev1v6qigBy22238dZbb+FwOEItil86Qgzm28ByfyeFEHcLIXKEEDnFxcXtKFb7IUwmwjIzcVVW+F8jBLNixzNtxwhMl6l30yIjMcQnYDt+jKo12diOH8d+7BjS6WxUFtAQWVeHIT7eq+lUQ1TuS8ciMzOTjIwMVq3yLiXpKARNwQghVgkh9vn4LGyw5ueAE3jD332klC9IKbOklFnJycnBEjfkmPr08dlO81LuuP56Xv7gfZx+rJ0LCIMBY2IShpgYrIcOYT18BIMf6wVAr61xy+CHsrIyVq9ezeLFi5v+RRTtyu23385rr70WajH8EjQFI6WcJ6Uc4ePzIYAQ4k7gOuA22dFD4e2AFhaGOT29SSsGYPywYfRKTml2+0RhNGJMSsaUltZ01ayUGBMS/Z7+73//y9VXX01sgDueKdrGzTffzKeffkpV/aC+jkaodpHmA48AN0gpgz+ysJNg7tsPHM7Lbj3ee/PN/OO/bwfsubrdjgiPQIv0nUau6zp/+ctfuOeeewL2TEVgSEpKYu7cufzrX/8KtSg+CVUM5jkgGlgphNglhPhniOToUBiiIjGmpfncUWrIDbNmkXv2LLsOHQrIc/Xqasy9e/stgFu2bBkRERHMnt30dAFFaHj44Yd59tlnL+s2h4JQ7SJlSin7SCnH1H/Uq7Eey4CMRjNxfGEyGrn7q18NnBXjcmFM8R/feuaZZ3j44YfVWNgOyuTJk+nTpw/vvPNOqEXxoiPsIikaYIiLc/dhvYxPfefCr/Dp+vUUlpa26XnS5QKDoVHbhoZs2bKF06dP89WvfrVNz1EEl5/+9Kc89dRTHS6zVymYDoglM/OyVkxibCyL5s7lpfffb9Oz9JoaTD17+g0AP/PMM/z4xz9WfV86OAsWLMDhcHS4LWulYDoghrg4jD17XtaKufdrN7P0vXfbNFJUWq0YU3v6PHf06FHWrVvHt7/97VbfX9E+aJrGI488wtNPPx1qURqhFEwHxZKZibTWNWnFDMvIYPiATN5t5VtL1jcOMdZXVV/KH//4R773ve8RFRXl87yiY3HLLbdw6NAhduzYEWpRPCgF00FprhVz3y0385c33sDVwpaZ4LZeDPFxPnuzFhUV8dZbb3H//fe3+L6K0GA2m/nxj3/MM888E2pRPCgF04EJa4YVc/WUqcTHxPDCu++2+P56TY3f1pjPPfccX/va1+jRo0eL76sIHd/97ndZtWqVpxlVqFEKpgNjjI/H2KNHk1aMEII/P/IIv3tpKWeLilr2ACnrR9k25sSJE/zjH//g4YcfbqnIihATHR3N97//fR59tGM0KFAKpoMTNnAgurXOb6U1wJD0dL6zaDE//cufm31f6XAgwsxo0dGNjuu6zre+9S0ee+wxMjMzWyu2IoQ8+uij7Nmzh7ffDly2d2tRCqaDY4yPJ3z0aJwlxU2OJnnkzjvZffgwKzZtatZ99ZoaTD7GV/z1r3/F5XLxwAMPtEluRegIDw/n3//+Nz/84Q85G4Dxwm1BKZhOQFjfvoSPHImzpMRvtXW4xcIfH3qYn/zhGWqt1sveUzocmFJSGh07cuQITz75JK+++iqGII4TVQSfCRMmcM8993DXXXeFNPlOKZhOgjk9HcvgQbiKi/wqmXmTJzN+2DCefuWVJu8l7XaE2exp+g3gcrm48847+dWvfqVcoy7CL37xCwoLC1m6dGmLr3W5XJS3cmJoQ5SC6SQIIQgbNAhzxgCcxcV+30pP/ejHvPrRhz4nQV7Adb6SsEEDPUPXpJTc/fCvCQsL47777guK/Ir2x2Qy8dprr/HYY4+1aFeprq6Om266iV//+tdtlkEpmE6EEALLsKGY+/fza8n0TErinT88ywNPP8U/fAT5pNMJmgFz/fa0lJKfPP4Mr/7zr9z186fUrKMuxrBhw3j00UdZvHgxx48fv+z60tJS5s6dS3h4eEDyadS3qZMhNI3wESMwZ2bi9KNksoYPZ/ULL/LyB+/z0LPPNkrCc1WUYxk8CGE2U1JSwqJFi1i6dCk9b3ual3fXdrhiOUXb+dGPfsS3vvUtJk+ezIsvvuj37/jkyZNMmTKF6dOn89prr2FuYvhec1EKphMiNA3LkCGEDx2Ks6jI5xZ2v7Q0Vr3wIodOnuSWRx7hwIkTFBQWYne5MPXuzZo1axgzZgzmhFR63fEHTIm9OVdpJftwC3NpFB0eTdN44IEH+OKLL3j++ee54YYbKCwsBKCqqoojR47w8ccfM23aNO6//36eeipwlqya7NjJsZ08iXXvXgwJiQiTyeu8w+nk0b/8hTXbtlFeWUF5dTUWi4WYmBhefvllnt4XxvHii4PfBiRHsuonM1Xvly6K3W7n8ccf57nnnkPXdVwuF6mpqaSmpvLggw+yaNGiVt3X32RHpWC6APb8fOp27kSLivY7FUDqOq6yUqLmzKHG4cBkMrE59zw/eHMntfaLLlSE2cBzXx/LnCGqRKArU1RURFhYGDExMQF5majRsV0Yc69eRE6dirTbfQ5cA3fsxZyejiE8nJiYGCwWC0uWHWykXABq7S6WLDuoYjFdnJSUFGJjY4NuqSoF00UwxscTNX0aWmSEO+tXSqTDgV5Tg7O8HFwuwvr186zPPlzEuUrfCXkqFqMIFErBdCG08HAiJ03C1Ks3rpJidJsNLSaasPR0IidN8oyMlVL6tF4uoKwYRaBQfRC7GMJkImLMaOTIEX7bYK47WsKJ4hq0JqzjE8U1rDtawsxBXXfYnSL4KAXTRWlqyFq/hAgeunrwZe/RL8H3nCSFormERMEIIX4DLAR0oAi4U0oZ2rLPbkT/pEjum63qjRTBJ1QxmGeklKOklGOAT4D/C5EcCoUiiIRq8FrD0YWRgIomKhRdkJDFYIQQS4A7gErA70xSIcTdwN0Affv2bR/hFApFQAhaJq8QYhXga+DOz6WUHzZY9yhgkVL+6nL3VJm8CkXHxF8mb9AsGCnlvGYufQP4FLisglEoFJ2LkMRghBADG/y4EDgUCjkUCkVwCVUM5vdCiMG4t6lzgXtCJIdCoQginaqaWghRjFshhYIkoCREz24JSs7AouRsHv2klF5p351KwYQSIUSOryBWR0PJGViUnG1DFTsqFIqgoRSMQqEIGkrBNJ8XQi1AM1FyBhYlZxtQMRiFQhE0lAWjUCiChlIwCoUiaCgF0wqEEA8KIaQQIinUsvhCCPGMEOKQEGKPEOJ9IURcqGVqiBBivhDisBDimBDiZ6GWxxdCiD5CiGwhxAEhxH4hxAOhlskfQgiDEGKnEOKTUMtyKUrBtBAhRB/gKiAv1LI0wUpghJRyFHAEeDTE8ngQQhiAvwMLgGHArUKIYaGVyidO4EEp5TBgMnBfB5UT4AHgYKiF8IVSMC3nT8AjdOAeNlLKFVLKC+MetwC9QynPJUwEjkkpT0gp7cBbuOvROhRSynNSyi/r/78K9z/gXqGVyhshRG/gWmBpqGXxhVIwLUAIsRDIl1LuDrUsLeDbwPJQC9GAXsDpBj+foQP+w22IEKI/MBbYGmJRfPFn3C887yHlHQDV9PsSmupjAzyG2z0KOc3ptyOE+DluU/+N9pStKyGEiALeBX50SSfGkCOEuA4oklLuEELMCrE4PlEK5hL89bERQowE0oHd9dPwegNfCiEmSikL2lFE4PL9doQQdwLXAXNlx0p2ygf6NPi5d/2xDocQwoRbubwhpXwv1PL4YCpwgxDiGsACxAghXpdS3h5iuTyoRLtWIoQ4BWRJKTtcpa0QYj7wR2CmlLI41PI0RAhhxB14notbsWwHvi6l3B9SwS5BuN8i/wLKpJQ/CrE4l6XegnlISnldiEVphIrBdE2eA6KBlUKIXUKIf4ZaoAvUB59/AHyOO3D6346mXOqZCnwDmFP/Z7ir3lJQtABlwSgUiqChLBiFQhE0lIJRKBRBQykYhUIRNJSCUSgUQUMpGIVCETSUglEoFEFDKRiFQhE0lIJRtCtCiAn1fWosQojI+l4rI0ItlyI4qEQ7RbsjhHgSd+1MOHBGSvm7EIukCBJKwSjaHSGEGXcNkhWYIqV0hVgkRZBQLpIiFCQCUbjrpSwhlkURRJQFo2h3hBAf4e5klw6kSil/EGKRFEFC9YNRtCtCiDsAh5Tyzfr+vJuEEHOklGtCLZsi8CgLRqFQBA0Vg1EoFEFDKRiFQhE0lIJRKBRBQykYhUIRNJSCUSgUQUMpGIVCETSUglEoFEHj/wMeQ4ag134dRwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 288x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = torch.linspace(start=-5, end=5, steps=50, device=config['device'])\n",
    "\n",
    "temp = np.random.binomial(n=1, p=0.5)\n",
    "if temp == 1:\n",
    "    y = sine_generator.generate_label()\n",
    "else:\n",
    "    y = line_generator.generate_label()\n",
    "\n",
    "y_noisy = y + torch.randn_like(input=y) * sine_generator.noise_std\n",
    "\n",
    "# split data into train and validation\n",
    "split_data = config['train_val_split_function'](eps_data=[sine_generator.x[None, :], y_noisy[None, :]], k_shot=config['k_shot'])\n",
    "\n",
    "# move data to GPU (if there is a GPU)\n",
    "x_t = split_data['x_t'].to(config['device'])\n",
    "y_t = split_data['y_t'].to(config['device'])\n",
    "x_v = split_data['x_v'].to(config['device'])\n",
    "y_v = split_data['y_v'].to(config['device'])\n",
    "\n",
    "# MAML\n",
    "adapted_hyper_net = maml.adaptation(x=x_t, y=y_t, model=models['maml'])\n",
    "y_maml = maml.prediction(x=x[:, None], adapted_hyper_net=adapted_hyper_net, model=models['maml'])\n",
    "\n",
    "# Vampire2\n",
    "y_list = []\n",
    "for _ in range(config_vampire2['num_models']):\n",
    "    adapted_hyper_net = vampire2.adaptation(x=x_t, y=y_t, model=models['vampire2'])\n",
    "    y_vampire2 = vampire2.prediction(x=x[:, None], adapted_hyper_net=adapted_hyper_net, model=models['vampire2'])\n",
    "    y_vampire2 = torch.stack(tensors=y_vampire2, dim=0)\n",
    "    y_vampire2 = torch.squeeze(input=y_vampire2)\n",
    "    y_list.append(y_vampire2)\n",
    "\n",
    "y_vampire2 = torch.cat(tensors=y_list, dim=0)\n",
    "y_vampire2_mean = torch.mean(input=y_vampire2, dim=0).detach().cpu().numpy()\n",
    "y_vampire2_std = torch.std(input=y_vampire2, dim=0, unbiased=True).detach().cpu().numpy()\n",
    "\n",
    "# plot\n",
    "y = torch.squeeze(y)\n",
    "\n",
    "plt.figure(figsize=(4, 4))\n",
    "plt.scatter(x=x_t.cpu().numpy(), y=y_t.cpu().numpy(), s=80, marker='^', color='C0')\n",
    "plt.plot(x.cpu().numpy(), y, color='black', linewidth=1, linestyle='-')\n",
    "plt.plot(x.cpu().numpy(), y_maml.detach().cpu().numpy(), color='C2', linestyle='--')\n",
    "plt.fill_between(\n",
    "    x=x.cpu().numpy(),\n",
    "    y1=y_vampire2_mean + 3 * y_vampire2_std,\n",
    "    y2=y_vampire2_mean - 3 * y_vampire2_std,\n",
    "    color='C3',\n",
    "    alpha=0.25\n",
    ")\n",
    "plt.xlabel('x')\n",
    "plt.ylabel('y')\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}