recover_michaels_model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "8fcb1d27",
   "metadata": {},
   "outputs": [],
   "source": [
    "# The purpose of this notebook is to create a pre-recovered mRNN. Recovery is obviously tied to a particular lesion.\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import os\n",
    "import random\n",
    "\n",
    "import numpy as np\n",
    "import torch\n",
    "from torch.optim import Adam\n",
    "import scipy.io\n",
    "\n",
    "from experiment import lesion\n",
    "from experiment import mRNN\n",
    "from experiment import michaels_load\n",
    "from experiment import stim\n",
    "from experiment import utils\n",
    "\n",
    "# Relative to the cwd. This is where we will store the recovered model\n",
    "OUT_DIR = \"experiment/recovered\"\n",
    "\n",
    "CUDA = torch.device(0)\n",
    "#CUDA = None\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "a2466da7",
   "metadata": {},
   "outputs": [],
   "source": [
    "mike = mRNN.MichaelsRNN(init_data_path=michaels_load.get_default_path(),\n",
    "                       cuda=CUDA)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "8e30ef16",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Get the Michaels session data\n",
    "dataset = mRNN.MichaelsDataset(michaels_load.get_default_path(),\n",
    "    cuda=CUDA, with_label=True)\n",
    "\n",
    "# Dataset items are:\n",
    "#   (michaels network input, trial end indicator, trial len (scalar),\n",
    "#    target output, class label)\n",
    "\n",
    "dataloader = torch.utils.data.DataLoader(dataset, batch_size=len(dataset))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "39572c5a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input shape: torch.Size([502, 341, 21])\n",
      "Output shape: torch.Size([502, 341, 50])\n"
     ]
    }
   ],
   "source": [
    "# The whole dataset\n",
    "all_data = next(iter(dataloader))\n",
    "\n",
    "din = all_data[0]\n",
    "dout = all_data[-2]\n",
    "trial_end = all_data[1]\n",
    "\n",
    "batch_size = din.shape[0]\n",
    "trial_len = din.shape[1]\n",
    "out_dim = dout.shape[-1]\n",
    "\n",
    "dout = utils.trunc_to_trial_end(dout, trial_end)\n",
    "\n",
    "\n",
    "# Batch idx, time, feature\n",
    "print(\"Input shape:\", din.shape)\n",
    "\n",
    "# Batch idx, time, muscle idx\n",
    "print(\"Output shape:\", dout.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "faead46b",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Healthy loss: 0.0005663853953592479\n"
     ]
    }
   ],
   "source": [
    "# Ground truth vs. un-lesioned network. We use this to measure how much recovery we've seen.\n",
    "\n",
    "preds = torch.empty(dout.shape)\n",
    "if CUDA:\n",
    "    preds = preds.cuda(CUDA)\n",
    "\n",
    "mike.reset()\n",
    "\n",
    "for tidx in range(trial_len):\n",
    "    mike_in = din[:, tidx, :].T\n",
    "    mike_out = mike(mike_in)\n",
    "    preds[:, tidx, :] = mike_out\n",
    "\n",
    "healthy_preds = utils.trunc_to_trial_end(preds, trial_end)\n",
    "\n",
    "healthy_loss = torch.nn.MSELoss()(healthy_preds, dout)\n",
    "print(\"Healthy loss:\", healthy_loss.item())\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "5374e3be",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Now set up our lesion.  If we want to try different lesions, change the lesion def here.\n",
    "connections = [\n",
    "        # No F5->M1\n",
    "        (1, 2, 0, 1),\n",
    "        # No M1->F5\n",
    "        (0, 1, 1, 2),\n",
    "        # For good measure: no AIP->M1\n",
    "        (2, 3, 0, 1),\n",
    "        # For good measure: no M1->AIP\n",
    "        (0, 1, 2, 3)\n",
    "]\n",
    "mike.reset()\n",
    "lesion_instance = lesion.LesionConnectionsByIdxs(100, connections, cuda=CUDA)\n",
    "#lesion_instance = lesion.LesionOutputsByIdxs(100, 50, 100, cuda=CUDA)\n",
    "mike.set_lesion(lesion_instance)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "32eaabb6",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Lesioned loss: 0.020719226449728012\n"
     ]
    }
   ],
   "source": [
    "# Ground truth vs. lesioned network.\n",
    "\n",
    "preds = torch.empty(dout.shape)\n",
    "if CUDA:\n",
    "    preds = preds.cuda(CUDA)\n",
    "\n",
    "mike.reset()\n",
    "\n",
    "for tidx in range(trial_len):\n",
    "    mike_in = din[:, tidx, :].T\n",
    "    mike_out = mike(mike_in)\n",
    "    preds[:, tidx, :] = mike_out\n",
    "\n",
    "lesioned_preds = utils.trunc_to_trial_end(preds, trial_end)\n",
    "\n",
    "lesioned_loss = torch.nn.MSELoss()(lesioned_preds, dout)\n",
    "print(\"Lesioned loss:\", lesioned_loss.item())\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "516f7190",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model will be written to: experiment/recovered/recovered_mrnn_connectionsIdxs[(1, 2, 0, 1), (0, 1, 1, 2), (2, 3, 0, 1), (0, 1, 2, 3)].model\n",
      "Loss: 0.02072\n",
      "Loss: 0.01968\n",
      "Loss: 0.01866\n",
      "Loss: 0.01767\n",
      "Loss: 0.01671\n",
      "Loss: 0.01579\n",
      "Loss: 0.01492\n",
      "Loss: 0.01407\n",
      "Loss: 0.01326\n",
      "Loss: 0.01250\n",
      "Loss: 0.01177\n",
      "Loss: 0.01110\n",
      "Loss: 0.01046\n",
      "Loss: 0.00989\n",
      "Loss: 0.00935\n",
      "Loss: 0.00884\n",
      "Loss: 0.00839\n",
      "Loss: 0.00797\n",
      "Loss: 0.00760\n",
      "Loss: 0.00726\n",
      "Loss: 0.00695\n",
      "Loss: 0.00667\n",
      "Loss: 0.00642\n",
      "Loss: 0.00620\n",
      "Loss: 0.00600\n",
      "Loss: 0.00583\n",
      "Loss: 0.00568\n",
      "Loss: 0.00554\n",
      "Loss: 0.00541\n",
      "Loss: 0.00530\n",
      "Loss: 0.00521\n",
      "Loss: 0.00514\n",
      "Loss: 0.00508\n",
      "Loss: 0.00503\n",
      "Loss: 0.00499\n",
      "Loss: 0.00495\n",
      "Loss: 0.00491\n",
      "Loss: 0.00487\n",
      "Loss: 0.00483\n",
      "Loss: 0.00479\n",
      "Loss: 0.00476\n",
      "Loss: 0.00472\n",
      "Loss: 0.00468\n",
      "Loss: 0.00465\n",
      "Loss: 0.00462\n",
      "Loss: 0.00459\n",
      "Loss: 0.00456\n",
      "Loss: 0.00454\n",
      "Loss: 0.00451\n",
      "Loss: 0.00448\n",
      "Loss: 0.00446\n",
      "Loss: 0.00444\n",
      "Loss: 0.00441\n",
      "Loss: 0.00439\n",
      "Loss: 0.00437\n",
      "Loss: 0.00435\n",
      "Loss: 0.00433\n",
      "Loss: 0.00432\n",
      "Loss: 0.00430\n",
      "Loss: 0.00429\n",
      "Loss: 0.00428\n",
      "Loss: 0.00426\n",
      "Loss: 0.00425\n",
      "Loss: 0.00424\n",
      "Loss: 0.00423\n",
      "Loss: 0.00423\n",
      "Loss: 0.00422\n",
      "Loss: 0.00421\n",
      "Loss: 0.00420\n",
      "Loss: 0.00419\n",
      "Loss: 0.00419\n",
      "Loss: 0.00418\n",
      "Loss: 0.00417\n",
      "Loss: 0.00417\n",
      "Loss: 0.00416\n",
      "Loss: 0.00415\n",
      "Loss: 0.00415\n",
      "Loss: 0.00414\n",
      "Loss: 0.00414\n",
      "Loss: 0.00413\n",
      "Loss: 0.00412\n",
      "Loss: 0.00412\n",
      "Loss: 0.00412\n",
      "Loss: 0.00411\n",
      "Loss: 0.00411\n",
      "Loss: 0.00410\n",
      "Loss: 0.00410\n",
      "Loss: 0.00409\n",
      "Loss: 0.00409\n",
      "Loss: 0.00408\n",
      "Loss: 0.00408\n",
      "Loss: 0.00408\n",
      "Loss: 0.00407\n",
      "Loss: 0.00407\n",
      "Loss: 0.00407\n",
      "Loss: 0.00406\n",
      "Loss: 0.00406\n",
      "Loss: 0.00406\n",
      "Loss: 0.00405\n",
      "Loss: 0.00405\n",
      "Loss: 0.00405\n",
      "Loss: 0.00405\n",
      "Loss: 0.00404\n",
      "Loss: 0.00404\n",
      "Loss: 0.00404\n",
      "Loss: 0.00404\n",
      "Loss: 0.00403\n",
      "Loss: 0.00403\n",
      "Loss: 0.00403\n",
      "Loss: 0.00403\n",
      "Loss: 0.00402\n",
      "Loss: 0.00402\n",
      "Loss: 0.00402\n",
      "Loss: 0.00402\n",
      "Loss: 0.00402\n",
      "Loss: 0.00401\n",
      "Loss: 0.00401\n",
      "Loss: 0.00401\n",
      "Loss: 0.00401\n",
      "Loss: 0.00401\n",
      "Loss: 0.00401\n",
      "Loss: 0.00400\n",
      "Loss: 0.00400\n",
      "Loss: 0.00400\n",
      "Loss: 0.00400\n",
      "Loss: 0.00400\n",
      "Loss: 0.00400\n",
      "Loss: 0.00399\n",
      "Loss: 0.00399\n",
      "Loss: 0.00399\n",
      "Loss: 0.00399\n",
      "Loss: 0.00398\n",
      "Loss: 0.00398\n",
      "Loss: 0.00398\n",
      "Loss: 0.00398\n",
      "Loss: 0.00398\n",
      "Loss: 0.00397\n",
      "Loss: 0.00397\n",
      "Loss: 0.00397\n",
      "Loss: 0.00397\n",
      "Loss: 0.00396\n",
      "Loss: 0.00396\n",
      "Loss: 0.00396\n",
      "Loss: 0.00396\n",
      "Loss: 0.00396\n",
      "Loss: 0.00395\n",
      "Loss: 0.00395\n",
      "Loss: 0.00395\n",
      "Loss: 0.00395\n",
      "Loss: 0.00395\n",
      "Loss: 0.00394\n",
      "Loss: 0.00394\n",
      "Loss: 0.00394\n",
      "Loss: 0.00394\n",
      "Loss: 0.00394\n",
      "Loss: 0.00394\n",
      "Loss: 0.00394\n",
      "Loss: 0.00393\n",
      "Loss: 0.00393\n",
      "Loss: 0.00393\n",
      "Loss: 0.00393\n",
      "Loss: 0.00393\n",
      "Loss: 0.00393\n",
      "Loss: 0.00393\n",
      "Loss: 0.00393\n",
      "Loss: 0.00392\n",
      "Loss: 0.00392\n",
      "Loss: 0.00392\n",
      "Loss: 0.00392\n",
      "Loss: 0.00392\n",
      "Loss: 0.00392\n",
      "Loss: 0.00392\n",
      "Loss: 0.00392\n",
      "Loss: 0.00392\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00391\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00390\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00389\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00388\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00387\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00386\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00385\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00385\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00385\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00383\n",
      "Loss: 0.00384\n",
      "Loss: 0.00384\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n",
      "Loss: 0.00383\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-8-c7fffbf8f2aa>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m     29\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     30\u001b[0m     \u001b[0mlesioned_loss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnn\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mMSELoss\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpreds\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdout\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 31\u001b[0;31m     \u001b[0mlesioned_loss\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbackward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     32\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     33\u001b[0m     \u001b[0;31m# This ensures we don't break some of the modularity invariants, basically.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.8/site-packages/torch/_tensor.py\u001b[0m in \u001b[0;36mbackward\u001b[0;34m(self, gradient, retain_graph, create_graph, inputs)\u001b[0m\n\u001b[1;32m    253\u001b[0m                 \u001b[0mcreate_graph\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcreate_graph\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    254\u001b[0m                 inputs=inputs)\n\u001b[0;32m--> 255\u001b[0;31m         \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mautograd\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbackward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mgradient\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mretain_graph\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcreate_graph\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minputs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    256\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    257\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mregister_hook\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mhook\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.local/lib/python3.8/site-packages/torch/autograd/__init__.py\u001b[0m in \u001b[0;36mbackward\u001b[0;34m(tensors, grad_tensors, retain_graph, create_graph, grad_variables, inputs)\u001b[0m\n\u001b[1;32m    145\u001b[0m         \u001b[0mretain_graph\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcreate_graph\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    146\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 147\u001b[0;31m     Variable._execution_engine.run_backward(\n\u001b[0m\u001b[1;32m    148\u001b[0m         \u001b[0mtensors\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mgrad_tensors_\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mretain_graph\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcreate_graph\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    149\u001b[0m         allow_unreachable=True, accumulate_grad=True)  # allow_unreachable flag\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "# Now we do some recovery.\n",
    "model_out_path = os.path.join(OUT_DIR, \"recovered_mrnn_%s.model\" % str(lesion_instance))\n",
    "print(\"Model will be written to:\", model_out_path)\n",
    "\n",
    "mike.reset()\n",
    "mike_opt = Adam(mike.parameters(), lr=1e-4)\n",
    "\n",
    "cur_loss = lesioned_loss\n",
    "eidx = 0\n",
    "\n",
    "# We will presumably never hit this condition?\n",
    "while cur_loss > healthy_loss:\n",
    "    mike_opt.zero_grad()\n",
    "    \n",
    "    if (eidx % 100) == 0:\n",
    "        torch.save(mike.state_dict(), model_out_path)\n",
    "        \n",
    "    preds = torch.empty(dout.shape)\n",
    "    if CUDA:\n",
    "        preds = preds.cuda(CUDA)\n",
    "\n",
    "    mike.reset()\n",
    "\n",
    "    for tidx in range(trial_len):\n",
    "        mike_in = din[:, tidx, :].T\n",
    "        mike_out = mike(mike_in)\n",
    "        preds[:, tidx, :] = mike_out\n",
    "    preds = utils.trunc_to_trial_end(preds, trial_end)\n",
    "    \n",
    "    lesioned_loss = torch.nn.MSELoss()(preds, dout)\n",
    "    lesioned_loss.backward()\n",
    "    \n",
    "    # This ensures we don't break some of the modularity invariants, basically.\n",
    "    if lesion_instance.application == \"connection\":\n",
    "        mike.set_connection_coadap_grads()\n",
    "    else:\n",
    "        mike.set_end_to_end_coadap_grads()\n",
    "    mike_opt.step()\n",
    "    \n",
    "    cur_loss = lesioned_loss\n",
    "    \n",
    "    print(\"Loss: %0.5f\" % (cur_loss.item()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "0d0c648f",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Graph before/after to make sure we are doing what we think we are\n",
    "mike_orig = mRNN.MichaelsRNN(init_data_path=michaels_load.get_default_path(),\n",
    "                       cuda=CUDA)\n",
    "\n",
    "I_diff = mike_orig.I - mike.I\n",
    "J_diff = mike_orig.J - mike.J\n",
    "S_diff = mike_orig.S - mike.S\n",
    "B_diff = mike_orig.B - mike.B\n",
    "\n",
    "# No learning can get back here because of the lesion; this should be blank\n",
    "plt.pcolormesh(I_diff.cpu().detach().numpy())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "f23b1f72",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAD8CAYAAAB+UHOxAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Z1A+gAAAACXBIWXMAAAsTAAALEwEAmpwYAAA8c0lEQVR4nO29e9RlR3XY+dvnnHvv9+yXHq2Wuo0eCGFBQAhZwYZ4MBgwOCvCHjtLzixMvFiWM4EJnnjGwXEmtteys0hmbJaZ5ZCIwFgkxhi/QENIbB7K2DADsiSE0ANBo3erH6hb3d/7u/ecs+ePepw6995+qB/09/Xdv7Xu+s6tU6dq17ndtat27dolqophGIYxeWTnWwDDMAzj/GAKwDAMY0IxBWAYhjGhmAIwDMOYUEwBGIZhTCimAAzDMCaUkyoAEZkSkbtF5Osi8pCI/IZPv0pEvioie0Xkj0Sk69N7/vtef//Kc9wGwzAM4zQ4lRnAOvAGVX0lcAPwYyLyGuBfAx9Q1RcDzwPv8vnfBTzv0z/g8xmGYRgbjJMqAHUs+a8d/1HgDcCf+PQ7gLf761v8d/z9N4qInC2BDcMwjLNDcSqZRCQH7gVeDPwe8B3gqKqWPsszwBX++grgaQBVLUXkGHAR8NxQmbcBtwFIr/vqzmWXnFlLDMMwJoz+k/ueU9XT7jxPSQGoagXcICLbgD8HXnq6FSZl3g7cDtC7crde9i/ee6ZFGoZhTBRP/fwvP3kmz78gLyBVPQrcBfwgsE1EggLZDezz1/uAPQD+/lbg8JkIaRiGYZx9TsUL6BI/8kdEpoE3AY/gFMFP+WzvBD7tr+/03/H3v6gWcc4wDGPDcSomoF3AHX4dIAM+qaqfEZGHgU+IyG8CXwM+4vN/BPiPIrIXOALceg7kNgzDMM6QkyoAVX0AeNWY9MeAm8ekrwE/fVakMwzDMM4ZthPYMAxjQjEFYBiGMaGYAjAMw5hQTAEYhmFMKKYADMMwJhRTAIZhGBOKKQDDMIwJxRSAYRjGhGIKwDAMY0IxBWAYhjGhmAIwDMOYUEwBGIZhTCimAAzDMCYUUwCGYRgTiikAwzCMCcUUgGEYxoRiCsAwDGNCMQVgGIYxoZgCMAzDmFBMARiGYUwopgAMwzAmFFMAhmEYE4opAMMwjAnFFIBhGMaEYgrAMAxjQjEFYBiGMaGYAjAMw5hQTqoARGSPiNwlIg+LyEMi8l6f/usisk9E7veftyXP/IqI7BWRR0XkLeeyAYZhGMbpUZxCnhL4JVW9T0TmgXtF5HP+3gdU9f9IM4vI9cCtwMuAy4HPi8hLVLU6m4IbhmEYZ8ZJZwCqul9V7/PXi8AjwBUneOQW4BOquq6qjwN7gZvPhrCGYRjG2eMFrQGIyJXAq4Cv+qT3iMgDIvJREdnu064Ank4ee4YTKwzDMAzjPHDKCkBE5oA/BX5RVReADwHXADcA+4HffiEVi8htInKPiNxTLS6/kEcNwzCMs8ApKQAR6eA6/z9Q1T8DUNWDqlqpag18mMbMsw/Ykzy+26e1UNXbVfUmVb0pn589kzYYhmEYp8GpeAEJ8BHgEVX9nSR9V5LtJ4AH/fWdwK0i0hORq4BrgbvPnsiGYRjG2eBUvIBeC7wD+IaI3O/T/jnwMyJyA6DAE8AvAKjqQyLySeBhnAfRu80DyDAMY+NxUgWgql8CZMytz57gmd8CfusM5DIMwzDOMbYT2DAMY0IxBWAYhjGhmAIwDMOYUEwBGIZhTCimAAzDMCYUUwCGYRgTiikAwzCMCcUUgGEYxoRiCsAwDGNCMQVgGIYxoZgCMAzDmFBMARiGYUwopgAMwzAmFFMAhmEYE4opAMMwjAnFFIBhGMaEYgrAMAxjQjEFYBiGMaGYAjAMw5hQTAEYhmFMKKYADMMwJhRTAIZhGBOKKQDDMIwJxRSAYRjGhGIKwDAMY0IxBWAYhjGhmAIwDMOYUEwBGIZhTCgnVQAiskdE7hKRh0XkIRF5r0/fISKfE5Fv+7/bfbqIyAdFZK+IPCAiN57rRhiGYRgvnFOZAZTAL6nq9cBrgHeLyPXA+4AvqOq1wBf8d4C3Atf6z23Ah8661IZhGMYZc1IFoKr7VfU+f70IPAJcAdwC3OGz3QG83V/fAnxMHV8BtonIrrMtuGEYhnFmvKA1ABG5EngV8FVgp6ru97cOADv99RXA08ljz/i04bJuE5F7ROSeanH5hcptGIZhnCGnrABEZA74U+AXVXUhvaeqCugLqVhVb1fVm1T1pnx+9oU8ahiGYZwFTkkBiEgH1/n/gar+mU8+GEw7/u8hn74P2JM8vtunGYZhGBuIU/ECEuAjwCOq+jvJrTuBd/rrdwKfTtJ/1nsDvQY4lpiKDMMwjA1CcQp5Xgu8A/iGiNzv0/458H7gkyLyLuBJ4O/7e58F3gbsBVaAnzubAhuGYRhnh5MqAFX9EiDHuf3GMfkVePcZymUYhmGcY2wnsGEYxoRiCsAwDGNCMQVgGIYxoZgCMAzDmFBMARiGYUwopgAMwzAmFFMAhmEYE4opAMMwjAnFFIBhGMaEYgrAMAxjQjEFYBiGMaGYAjAMw5hQTAEYhmFMKKYADMMwJhRTAIZhGBOKKQDDMIwJxRSAYRjGhGIKwDAMY0IxBWAYhjGhmAIwDMOYUEwBGIZhTCimAAzDMCYUUwCGYRgTiikAwzCMCcUUgGEYxoRiCsAwDGNCMQVgGIYxoZxUAYjIR0XkkIg8mKT9uojsE5H7/edtyb1fEZG9IvKoiLzlXAluGIZhnBmnMgP4feDHxqR/QFVv8J/PAojI9cCtwMv8M/9WRPKzJaxhGIZx9jipAlDVvwKOnGJ5twCfUNV1VX0c2AvcfAbyGYZhGOeIM1kDeI+IPOBNRNt92hXA00meZ3zaCCJym4jcIyL3VIvLZyCGYRiGcTqcrgL4EHANcAOwH/jtF1qAqt6uqjep6k35/OxpimEYhmGcLqelAFT1oKpWqloDH6Yx8+wD9iRZd/s0wzAMY4NxWgpARHYlX38CCB5CdwK3ikhPRK4CrgXuPjMRDcMwjHNBcbIMIvKHwOuBi0XkGeDXgNeLyA2AAk8AvwCgqg+JyCeBh4ESeLeqVudEcsMwDOOMOKkCUNWfGZP8kRPk/y3gt85EKMMwDOPcYzuBDcMwJhRTAIZhGBOKKQDDMIwJxRSAYRjGhGIKwDAMY0IxBWAYhjGhmAIwDMOYUEwBGIZhTCimAAzDMCYUUwCGYRgTiikAwzCMCcUUgGEYxoRiCsAwDGNCMQVgGIYxoZgCMAzDmFBMARiGYUwopgAMwzAmFFMAhmEYE4opAMMwjAnFFIBhGMaEYgrAMAxjQjEFYBiGMaGYAjAMw5hQTAEYhmFMKKYADMMwJhRTAIZhGBOKKQDDMIwJ5aQKQEQ+KiKHROTBJG2HiHxORL7t/2736SIiHxSRvSLygIjceC6FNwzDME6fU5kB/D7wY0Np7wO+oKrXAl/w3wHeClzrP7cBHzo7YhqGYRhnm5MqAFX9K+DIUPItwB3++g7g7Un6x9TxFWCbiOw6S7IahmEYZ5HTXQPYqar7/fUBYKe/vgJ4Osn3jE8bQURuE5F7ROSeamEFUZC6+YC/rsRda5J2vI/Pky/lTT0KxdG8na+UWHZI6x0oWnmm9xUtWca2oe9kK47m8Tre82W30hSy1YzZxwpmHyuiLLFd61kjQ1/cJ5XXf7JVly9fyukcKegcKchXskYWL//sE00bpp4tmHq2aMkkNezc/TwAl3/f4fY7WnflyUBifVLDdbevut8q+T0Mw9icFGdagKqqSOgOXtBztwO3A/Su3K0qgAzlyQBc0Spp2omp5qqmDIFyW9XOkAVxG7HXLytbWVavaH8fh3bd8yPlA5qPvhIV0Oma5auTXtPLohnQS9K7OpInljPt8lVzFcM1B1mG5V+7fLQ9msHBZ7YD8OxTF7WHA14W7Sjaaep/9LbpkTIMw9icnO5/34PBtOP/HvLp+4A9Sb7dPu2EyMCPgosaKWqKBTeCz2cGdA47HbXtsgWXpy9xBDp78TKdLesU8322XLrE9I5VAHoHR/WadGqmdqySrWUU832Xpm4EvHP388xevIwoTO0vmL14mV3/LWfPld9FOscf4k7td/Vc9DcFve1rrXvF3GBkdJzPtNOuvvoAWy5dAuAV1z3lZFjPmLlopZHbzyykEv7Tj/77KHtxLCdbzZjaXzC1v6D73abNF11+jM5hNwPIl3LypZytDxdsfbggnxm0ZJq7ZDleZ6sZ2WoGonFGkS/lvOhOyGcHzcykVyG9auR5wzA2F6J68sG7iFwJfEZVX+6//+/AYVV9v4i8D9ihqr8sIj8OvAd4G/C3gQ+q6s0nK7/3fXt016++Fw2dreKGy6JIBlqJGwXXMjIaFgFV9xeG8uI6+Xwhp9zeHgFLKVAJ6ke62y5d5Oih+Vhe57kOg4t9Z1kPTU1iIQoq5Ms59bYBOmj0qeSKKi05VJwikgXXWdezlavPyyy5umsgW3ZKsJ6pkLUcna7cexGQtQydqimOFNRTfgbRq1GBfDmnmi+RQYb0hXrWPVccc3WW28qmPV4oUVdm7WcWZOrq6GosL3D9bx7i4V+91L9saclsGMb3lqd+/pfvVdWbTvf5U3ED/UPg/wOuE5FnRORdwPuBN4nIt4Ef9d8BPgs8BuwFPgz841OSQkCLuvU9oGE9IKQl/b+Iv6+u0x6HZko1pV6p+E8taKFot45pq+tdX6jL2z02VF9q+B6WU0Y7QQ11JXIgTjHpVI1O1a32obTK0PkSnS+d1oiKUVyZuctf99R1/L06mpyqudLlydSZZ7wM1UxNNVO3ZEKd4lGB6f3j/ylUs5Wbhfn2LNx42Wg7DcPYlJx0DUBVf+Y4t944Jq8C737BUoThcdrZgutg/WhVU3NKSEufr6X5no7Y1Y/y/YwiLibU0urQB8/OwHQNlYDA8jWD0c5yWOxKnNjiFk1TW/nIrMHXny0UVLNVzCO1t6OrIKW4jlwUVv1CdrdGM2lmP+rXF1SQi9bR5U67Pi+n5urs+KHdQbl00gYol36p4NAPVSxfXTbvoxaYqaDMkFx50U3PsPc7u0CU3b+0l2cevnrkHRiGsfk440Xgc0LaSQ+njettTrUHGs6XfA/mj5h0PLPPODGnT9EVRqXp/ENSMvCOC8cqjTJJZR6SqVrpjCycj82rjamrVZ4Kh36oan2Pl6UTTCtxnb+//5WHrhlukmEYm5SN4cOhibtjKY3rZ+omGdwSK4n5srWscefsS5OnbHqlOJmopHk2uFOuZU2d/aTMUnjDKx9uyzLOApTeGwy5gQ6kJUdsz6BZxA51Bplbn/XMfYLcA2muQ5sTT6OwuJwvNy6p3a3rTZv9ArskZiaphKv+pI7yt95ReG8KN1//WHwXj//dD8d84N6hYRibk43xv1fUjWQzjWYOlw5M+RFq6laZK+SNe6LmCh1tFoiHF4oV6PlF1FxhunLrDt1m5F4sZ7EOLZT/50t/a6ScsXIHZofcLKdqtNBWVs1cerCna1edqQUni3aaexTqPmHtIFe3TpJ4sKrS6owBar8mIArVszONPP3MfYbkP/Aav/YxV0Ytl8oN8MDnX+LSMuUH/9d/1LpXT4+6wBqGsTnYIAoAqIV8MXefZSdWtpqRH3ZG63zJuyUuZnF0jCjFsRwyJT/WPJeaVYrnC+cu+XiXfClz3490oVehuZIvZ+TLGYPtle+Ua/LFnGq+QvoZxYKfVaxlyNBoN9ZTCdnRxrguCvnhIrqwAtDPyJdztBKylZxsxW3YUt8pF88XcR0gW8ugFCjdO5F+5v7W4to4VTuXzeUczdV9giyzJd3vdlBxHkT5UuZmNf6dhfaAm1WtXVo5M85i4X6HsJ6RWNzWLi8pjubkx3IO/HdVlA0gX9iYVkTDME7OhlAAkrvRZTXlPsWy61ymDmWw2/n2Tz3n0qqZGs0UzZR8MUdzyFZy6imNG7P+0Q/dFcuu96yiGay+aEDdUwYX+cXdtZze/g7VbE01W/OtH/8QDDKy1Tza1aWGcs6vDWwp0S3tUX70/Z8tqbc1/vUqTs5ySzI6nqkoloRLvpLDpWvuUzdmq3J7yY77CqR2i9A6VaFTFeUWp6jqnrLrRYeppmuy5zvUMzU3vXovn3nL7/KZt/wu//R1f+HqXs8ZzNfkMwNe+4pvOc8fAZ2r0LmKcqYZ3WuvpnvULzbPlfQOFPQOuA5d/OxIuq4N1bRS+fZ0FjM6i07uetZmAIaxWTmlfQDnmt5Vu/Wy/+2fNAnJPgD3PbGlp2YZxe0TSJsQvHvCM6Lk0yXVatEsmAZ/+tSH3XvZSOlcRFt7Do6zICyVxIXbtCwJnjrDz/p9DeReqfiF1li3NG0JZh3NFSkULWVE7qxXUa8WNJUKxXyfcrnTeOd4k1pYL9Bamr0JuUJeo/0c6VWtfQz5VOkXmTW6i6aytq5PYcHcMIyzz5nuA9gY83fvmz+aPiZtKJ+OG4AOebpUK52mnqROHa6zFmdKqROlcCKxk3WJlg9/4r46LJdWQJWP3kvdWNMm1IKmcYYSuWPnnzxQLnZHngecAhkWp5Ioi663ZWre2Zh3cbxrwzA2FRvCBBS9gDq1+yQ9YfRGCQu2oo1HS1E775dMkW4VwzYMBzyT2heUuWcRbQLO5W4H7tVXH0C6tSunW7vgcJ0mwty4wHDhe7GQR1NJzJvIE9ohNWTTZRNwragbmYNcndrJNPCeP7n68uq4SBvbvJ54MdVJPf69SNgVnTXlpTMoqeGtr3rArVmsJB5H3cZbSCph5snCyZcp3/efpfUupDv0UgzD2DRsEAUA2dwAlgr38S6VneddvJuQB3AhDpYKZKmAfobm6sImrBaIN62krokyU5Jv7VMcK5y9fT1nbm8HenUr5s7T916BHCtgsUN2tOAn/96XkcUCCZ4zU7X7JIR6qrmKYn+vudFRsiPd1sIw3rMmf2aK7o41ujvWYDWns9+P2MvMhWvwG8C0WzsvpYE4T6jVnKs/6V9Ct0bWM6YPZmR9IesLePNNZ9s6+XMdZp4q6B10C8jZck62rU+2rd+0x8v5xf9yIyrQ3d3E9JFjRXzvxbEcfdUirOTIas6+v993G8z8voLCy28YxuZjY6wBvGiPXvZr/1PjzuhdJtMlgGBvT2cHKmF07xNqnyfsrgU3+l3NIRmpxuWBTt24RpYChTYbhnNt7PDHsXIE83i+mFNvHzQ2fX9T6mSXrP9Ot246+UJjITFWUBAzmFbyofWIYC7L3UJ46gKqmVNK9ZSbReVb+lQLbZNQa3mkFN580zf4y6+9nEv3PM+hJ3fEd5b+Bt1DBf1L3QL4y657moe+uSfWJ0XdbrdhGN8zznksoO8JNa5TCz7wwV49kDgbIO0YQz6SvIDMOk+c4FkT8muvdn8F12muJ4uv3Rq6tYt+WbsNYShse2DMusEQ2YrryItVaUI3BFnKbHQncynIQuHiEBXa2pilufq6/SJ02CsgzVrD7KXL8VkqoXO4E11nQ5tC5z/37YLiWzPRZfPyL2Zc/sUhN9ZuzV/c/3JUXFjobF3I1t0icfe5IrZx7mmiPM99+MrWDGBkb4FhGJuGjbEInG7oAmcuCR13WFAtwiatxAwzHOLAe/rUs9XoYnF4TmieKSXOFKodJVqDdEFruPqn93LfN1/UlD2Gat5F1pSXLsLhoTj5ubaH2/iNZ0XyvcwaFayJXAoy75SZ9nO/OK0sPTfTKqu8tGrOExhksS4VWHzpgKxXoWs5ksGRn3Iuq/r8VCOTulDWh3/Aje7reb8prYb1y5zSrOZLircfRfdtA2B9q7QXlG0N2DA2LRtDAcTTs4JtwptDMlwnnfY3YQYA3hRCdI+MJJ2/hNlC7Xb6RvOHn3VEz1C/SKveVHPfI1e2TVBjPFID64en3YjYj4ajmUqHZK+cwkkXk4PLabwXPCzXikZO/1KceUuaWVCuTkEkhIVfrYV6zW0eo4L155yCEklMQJVw5DUDF46iljjLkiC7Nz8dfGZ7lOv5G8pm1pL5TWNDayOGYWwONoYCEB17spT6yJyQDFqFprMPdvETjEJj/uFTtbLjpCV2+yFv0lGCi6XQMoUcb80gtLHV1lra92T8M+GmCk1YjMHoS2udRBbyn0ie8I5HTjAbejfpu0gtbNb5G8amZWMZcEMsoDQS2XB8n5CWuHQG19DjxQKKacmzQHSnJFPn0ZM197PVrC3LOIKL6EBGTg5ryZPkT90mpaibgG5J3U20ON82dfdFab+b8HcoUp1481N0+xRtnfaViv/P/s5n2+0RHfkNpnesxrTpfUWrPilMARjGZmVjzADUd6DrzTBYc3ULqTQeOcEbqHkuQ0LwuEriKFb6WQxoJgoyXbq4+akdZ6pybqMr7hXofIms5dGcNP9YxsL316GaGGFTk7N6ZeDqyfqZs550mjpZzxGaEXm2lrk1gPUMmfMhJdbyuL4RzTDiCwh29kJhunKeTL79MluiSx1vsmm8hTRz+w9YLRBRsoFQ587mk/nFdK2kNYL/4303OvlWs6Zt/j1p5hazu1+aZ+2lTuZ8lZYJiLU8rs8YhrG52BgzAPU7VYO5Ynh0O8w4s4749YFwPyRPl2QHeu28vcrtIVhKT0fx5fm4RHWR1AMwV7nPGIpFaUUDDQHa0jp1vnTKYlufzr4enX09V7aPBuquSz/SHqp7xYexCPprwSmzztE8bgQLB75kR7qujEroHs7i+yyWhGJp2L4Eb7/864A/0yDOJogKSPrC8s2rUZbL7m6ffTwcOdQwjM3DxpgBZN5bJnj6KI1Pv7dThwXJVohlYfRIwlpadvB6rUAvKlv5ZJA5d9BO3YSDGGRIUv/idWUr5o2OcXcMsvR3lshCp1kMDX77QuyYdeDML9VKh2rnoJG/38xy6OfNLEbTOsIXXJm+3sEV/RG56m2DaPtf3VPGGENrV/hgdcGrypd9+6Ovcz/BahYDu4ngjqEs3TnB9VInyrX3HxSt95JNl+2QFIZhbBo2xgwgOfZRQwgDcB1ZGqwNmj0A6vK29gWEIGdDh7NcdNmx9nOZ96YZZDGtOJa78nzQtflHm3NwYfxkJJiFiiMF9czQ7CDIF/KGcnIXnC6fdm6nrQ1e/jxi1JmxpJ817dKm3JBWHOyQL+XkS3k8WCbrVe4dDoS5va4NWgm9Ax16BzrDa7usPjnvXu9MFWcTWuNMO/5d5st5rPf6f3Ww9V7qtTFxjQzD2BRsDAXgFxjj4SZppxtcE8PfStqfcDpWmTWKI22VKM/t39p6ppgtnVdL0dQ5u8/X7V0yl17eHzlsZQRv8x9cPGinp23xRG8cUaqVgmqlcDKHtYpKkLqRsZ6pXKdc+RPEkjaKP0u43Nmnmi/dfgSvSOr13NVbKKuvWI3vqL+nT39Pv30iWCnUuV/37VXR7CRVM9PSbs3sU41c7/9vf9Rqm9guYMPYtGyM/70qMQ7+cV06Zeh6OG/i3dMOGS3u0JJ07TiM5pOYQZ03PdesKwhk3ar9jD+DYBzFsYJ8OjkrQPwC6ZB8Ugm9LevkSwX5UjHqZdSt4zMyyJDg5hnXNnze0ImvFO3OWEEWm/UC5w3ky17NR3YrA1xz/T538Xy3Gdkni+1SCYuv7MdyfvKP/+fW8zriPmoYxmZhYxhvg9fJ6Bpu/NLE1GluxQ1aIU9YSxgqu5ovmwVioF5yJ2ZJCLkAHD60pRUvp3UsZUuAdv0A9aXr1CvJq6xldO3am3HWnp+C+bJJC2GYxa0TSKwzWUBOYyCps91L8jeVR6fr+E6qtSKWHUNpDCnSJw5e7J6bqeI7cpvwNObRfhYfK7dWySzCKzA7Gd4wNiUbYwYguJg/fhQfRtqa6cgIMx6B6H3d1QdFa8XmTw9L72fkM2XruVagN/89nyrjiFxz5WV79rdG/ePCQYfe97JLj7UPaPeL1uNkDwe6hFPQ4kEtedN2zTRGH9XC5Qnp7oCYuvEy8mcHtw6m8fkRmvfjYwi13lOm1M+5KKa9g52mbtFmQT79LZJwHGloDcMwNicbQwHU4nzQfZgD8VJJJdEMkk01o+Zh+3oIjxDt6ckOWe0o1XInLpICyGre6iQRqA73XBkdF6DtW3dd7YfUXlFkQ7tyk3r2H9zmTC+hzlydCWdo/UAqIV/K0UHmY/cAq42ZR1ab2QArufsorYB3CC6GkOICwPmAeaF96tcxAFj0JqIMrvy0+wxvb37dDzwCwPqlZVRMUjeL77Ke8f3/50J8x9/3GeK5DUFuwzA2JxtDAQwfujLm9KrR2Dg0Zp1gSg8ulcO7cnuVW2fwikOnKr/ALNHzZebyJVf2uut01y+p2qP+cW5AIXjbctsLKO7YTbP7mUW9pSSbdh+pJG6+khp0uooLwSHiposRpF7WLHb0IV8c2XtzUG/7mpNZBcKGs4Hw5E/XPPnTdfv9Kfz1117qmhbOFBgkygW38e2b/3TOpdXwc//m0+4dedPVyKzIMIxNw8ZQAAE/Gm+ZlIdDDgyb+DvNQmc016Qjb1G3+CnNaF/6WTSnhJH9JR+ZjSYTxM0cWgvNOiwYcefyzNO520WcZB0N6uM67GyhoF51H4TGZVVA/KK0ii973W/kCmVlPmZSWDOZrZh+ssP0kx23t0Ggf2SqeR9+hqKZO7il2N8dWUP5l2/4VPSIakxAkG1fd2KV4vcwOBl/469uGfsqDMPYfGwMBeDdI0PH0hqlJjtsIbFpB9u0jxba6nR7ozt2tWgigWpn1FT0zP8waAWBm95XtGLxj50BeNlWL6ubUM4tOYeE6CjVbNUuKqwBZIpO1Y2n0VTlPrTfTROmwRWytrtkbXcZbfKpp9KWS5eiLb+cqynn6nbHnSnfXN3l8u5aRDIQH1W68pu7tFuz664s1n/9rz3VrGGArQEYxibmjBSAiDwhIt8QkftF5B6ftkNEPici3/Z/t5+0IO/jni/n5Ms5xTE3ms639pHD/shBbx+X1Zx8sXCfY+6Ywmw1Q/oSR9NpiGQpM2Q9Y8tDHWfWCGsItbgDWvynXOpQHMt56TXPIusZq1eUrtzgKjpm2BsOUtep2m2WSt9NKe6QmZTSlSFrrtxsOY9KIsiWreTx6ErWc+hnSOkCub38JU+7c4D7GbKWkx3rjCi/rFeRLTlZFg/MwVrG1P4iLhZLcsB8tpLz9aNXALDw3BzZQkG2UFAcy92mMYBa2P8jNdIXstWMh3/j+8gWc7JFV8f8oxvDkcwwjBfO2ZgB/Iiq3pAcS/Y+4Auqei3wBf/9xPhRbdjUNLjIjazL5U48pCScx6vTFdWWkmpLSbmjhE5NPVu5MAwhFMPQSF0LZeFlA3S6Qju1WyPwWcLZu29/9X2U20sefXwXOu3qjOVyHC+gZBdvvaWZqUjtyq3m2gfFh1G7dtyn2tJs4NIpF4unnq1i/aG9QZYH9+52+TJgpmrNJmLoiGNdqq0lojB/2RLaq1nf3WxUS2P3a7dmz+xRtxHsWEE1V1LNlZTbS/qXlK1XqVN1DBVRzVWxbYsvG9oEZxjGpuFcmIBuAe7w13cAbz/pE+JcC0VwcWhiSIcsLgjHkBBK9BaK0TO9j32Tpxnlaq5ot47mHrej1kXEVL/IKpXwqXtvdN5I6c7WUEcoZ3jTU3Iv3VQWzhVI0Yw4i4jHOibluza4dojQbGxT3CHwWZNfi9qFe9BmYhLNV8nxl4sH5tx1JY3rZipXt+bzX3uZM/nMJ6a2uokYGl9leshO+J3S38UwjE3HmSoABf5SRO4Vkdt82k5V3e+vDwA7T60k31FVQ533cP8ybIaR8ekjZYdgctJ25xy7oDnOtfFEK58qcV3hpOWcjPAeEtOOVkOKKSiSkxQ/fJjMyLsrs/b7O9nqblB4Q7+TYRibkzNVAK9T1RuBtwLvFpEfTm+qarqntoWI3CYi94jIPdXCivPcGXe4ibdZi1/YTa07UiT+6Flz2ErL5z+NJeTNLdlqFmMIhZF2tNf7PFPPJlEvGW8CCmsOxTG3DtG6V7V3A0c51rOmrPSAeh8/KBx0k61lZGtZEw8pFOYPhhGFfDmPB73E8r1MUkPvQNHk9YfHt4TKlL934/3OffRQkxfRxt10IO6MAS/C9b/5bGsfwInOyzEMY2NzRgpAVff5v4eAPwduBg6KyC4A//fQcZ69XVVvUtWb8i0zZPP98XXEA9yDS2OjJLSW6OoINDFy0l25M87n3plgcM/1audhM10Ro2/uXHMdny9/fVcZ8wMuTk+3rQGCbFrQ3geQq1trSGMHdXzYhpmKbG5ANjeI5p1YXvDkqYR6qqb2XkFa+B3EtZM1eEHVnRrtqttLELxJtwxA3Aln1bUr0WtKLl9FLl8dernw+SdfgorbCJaauWLY7Qzy73bde8uUF//ZIed5lR5YYxjGpuS0FYCIzIrIfLgG3gw8CNwJvNNneyfw6VMpr1rpjDdDaGN2iN/DJzWxJLtX08frtaKJVx9MJ5nzFNJ+FtPKxa67V2aNeSXdCVxmbTOMrxOgmq3a6w5hN+5Qmor7W610YntHTDD1mPYH23+yjoC6TWQxvEVYSlgtQIVqpnZtqpv2hTamZa8cnmm3Z1juXCm3VrHOO++7of1ezBRkGJuWM5kB7AS+JCJfB+4G/rOq/lfg/cCbROTbwI/67yemFjrP5yNmFtHGnBP8zkMeCQdY9WX0ucTVcfaxgh33FU3o6BA+Orix+7SWGSWYZ5Jyi4WcYmHI1TOcwrWWjZxBEOoL5Mt5NOdc+uWcS7/s2xt39jZyidKYgHxs/3i/hnylSY8moEEii28PovE99bat0du2NvJ+t+5cdO/p8cadU2roHWx2Xt/55g/Gd/HS935j7DswDGPzcdpO3Kr6GPDKMemHgTe+oMIyZbC9ajrR4K0SbPQEN0U/Eg37oyph6lDG6hXuWSEcbdgUvXx1yTJeKWTOAydbFerpdiTNEP9GaqCGm17+OPfdd00MulZuGXMcZNiUtXWAHO20FoJlIGhOlL+arSgWcuqucui1rqxszYVd0ELd4S/gnqmbRWUtFDKvBHzoiGqmjuak6P3jn6+namQ940XX7+eZw9soj0yhhfIPXnoPAL//pb/TrG0MhJVvboPtFctXlc1u6+UCfckyHOuhhXLv2oucN1ahfPzbX+TV//crvKzq0i0ktGFsSjbGTuBa/KHkzic/oIVGN9B6q3NTdG6d7lPPVqzucWft1rMVGvzuk1Zlqxkv/njpXE27dfTDl9pbMqJLpr/fc3b1b955rbPrp4fLD/dzYQYQNqslaDhf2COKV2Bw7cf6XPuxPvVMFTv6eqp2+wN6tTsnuMJ9kv0K9UyFFsrUfrdLmeUcWSjcx8vSO1SgUxVPfGcnPOHCWyDKxz/1ej7+qde34iRpr+bHf/RvADdDCWYu7dUMDk85uQcZv/tv/3v3zjP4yV94b2s9RKfGKEbDMDYFoiOH6n7v6V25Wy/7F+8932IYhmFsKp76+V++N9mE+4LZGDMAwzAM43vOhlEA4g86kaKONuoYVplkETjErE++h4NVZIwtOpZbifubJ7ac4zw3XFfI2zpqMiVzbpdNAWPKleb5k9ZZ1NHm1Hovw+0f54Sf1lPKcd8b+H0UYeG7U7frKep2sek7T/IZhrF52TCRvJw74ZhdvkPuhsNuhzH9ZOXmGvcSRGoZ+9zYuk60q7d2fvup4DpsGvdhHsa1YbTO1H2Uoe8n2/6b1FNoXKcYW2eZxSFAaz9FuDdGvnH3DMPYnGys/8nDO4FlaAQ9nGfch7ZrotTJaDaMqsPu4mSkXSzkrXKm9hftMscMuKNIvao1A4juo+mOZu8Cmi/mzQg62XEbNm9FGbu129kcRvRj/sZZQFpPtzpum0Yakc4W1rOx+URh/lvNu7jqqoPHWRE3DGOzsTEUgB9dZys52UpOvuQmJvli0XTW3nslhEuWdRcSecslyxTPFz5ksvfTTz2JMjdizdYyOkcKiqNFs9FpkDH1bIepZztUl/Td8+LqHWytozxwkjA5Zdb2Xspc2cOHxPQOFtRzVfS2yZfz6PqaLxUuON26C/8cjoTsPNfxXlI51314nc7hTgwbnR8poozhPWk/J18o6B7qUM1VyHpGvlTEdubJ0ZUyyJh6xoV91l5NvlCQLxTIIKP7XZeerWb87D/8S1ffQsGzX70ivv9w3zCMzcnG+N8r/qDzKfcJYRXq2QqCb32Ivd+rXVpHoVuz+Ow81ZbKnZbVS0b6oeiidj753Zpya+Xyenu35sr6ZSXrl5VICBFRC/VMxStf/ZiTJ5Q5bg3A3wuxctL20GmHjpBcWd9ZtmxV9XRTfh3CUnR8PTMlzLjQzDIQtFfzrZ+botzmzinQqZpqvkK3lujW0rmOejnr2Yr+pQOy+QF0lHqmorpumeq65VbICgSufv0TAMw8WVDPVu6d50o56+SqZmr+0394M9pz4aDv+YcfiGcLAK3w0oZhbC421hpA2sGqOwu3OZTdXY+EZE6fEdxsot/YtulnMFuhyU7dbDWnzhXpaOyP80M9yrjXAB786xejlwyiLCGcdKt+f2SjPNdr/Pzxm8pCvjBtWHOzhHyxoN4+iG3OlnN02ikjWc+iQpDl5kQuCvVxiXAbx2adspBa3EyBJnaPFDXqw4AWj0+zvtPVVR5xfv3uLAT/TouabuZkXttZt9qmM014i6UfWEVXnDyvuPOfNHsJhn4jwzA2FxtjBuBj3LhDz7O2h0ywUYczAsKJXpUzi3S39JGBOyUrC6dy9YZMQP3MPTNwf7XjO62VPJ7OxeVrLk8GUma85Icej/mB8ecB+Hrmrj7qdiCHOoXWs4DbfLaWuw1rgwwdZNG8BE4p0fGHvw8yd7jNVN20t3Ttk0EWTWb5/IB8W598Wz+ayLTvTGRveMUj5N+/4N5rBu947Zd5x2u/7MxLHulnPPzX1wBuI1rWq8h6FVIJcxevuDy9ii3/77T7XSph9vE8vv9QhmEYm5ON8b9XaXf63twjg6xZJA2daTw4xY2yB0d7bmQcdvSCG/UnbLl0qXGkqcUdsOIjdpIBGXS6pSvXL8Q+/NQu4uEzMH7h09vBl1d6I+cQpztuw/Pqz/nNZwbkMwO3u9efsqU9PwKvvBdPqE9Bpv3MxJtb8tkBMsiojnYplzqUS53Y5t62Negod91/PatLPbcYvZ7xiW++mk9889VtGbs1L/qvTYRQXSncSF+F5UOzLnE1Z/lyL48KM6//bntBxKKBGsamZeOYgBRnNw/U0iyshutaWh2riH8uKI/k0JdIpiw8N+tmEtEsQ1Q66h3hV49Mu/prZ4rStaKZSRzHBTTUUy53nKIKekcUfJiJ1rMCup5TZT5jcj8eWNMdqrNQd8ZxOMYyV3dge9fH5F9vFqkB+kvdWJas5V4pKv1FH64itNHL8+13uMVeKWo0yqruxK/KRSDtXxKOrlQO7dve/E7D79owjE3FBpkBuIPSm3DEPj05MlHCSDM9DrKUOEqXKnkuGalnK7kzDWmS1x/OTr+ps7d9LXb+Ugr5Uh4VBThTx4i5w9eTLxQxZhF4M9V61ur8Zd0FfpO+s/tny3k8kD62i6S+5MD6aPYJR0aGdq8k+jsEpiu9nKJRBqmEfLFwHkDpkkkldJ9zZWQLRdPeJLQ2wMVXHI3vaetD7THDcBRUwzA2DxtDAYh6U8yQp433DgJitMuYJ/M2eZXmegwqOC+ZUG6m0HPRNEOkTzKlXM+bA9oLdQu6o3vTxlLvGMBc+0xdpuqRBepsNXdB1eZc4DottOW9I8lCbwi4poXzdortFlzZQoyKCs0aST7rTEtSCTrv2q2ZUs2X7tzfpD2aKTtvcqd3ttw5M21s/JVQfPyiWP/fuvXhlmLEvIAMY9OyMRQAfiQZOpa0kyqTjh+aPOlof+g5SUbe2UDoPNtt3c+e70CykAkg+6dao+N4ROTQiLxFsH8sFmRHO016ps47J13LHrjdwvlyTu+ZDj3vfy8L4bAamlmE4Ew73rwT1hrCQeys+rKn62apILTtqWlXXK7khzux3cWRguLIqMXv2IrLP/8UjcJTmtPJipoDP1LGd/zgH17ffgV2HoBhbFo2hgKoXZz7sKkpePPMXbwSO+Kw8UjWs2bD2GLhvH9W89bmpNaJYDMVg/mafKlwHj/rWRzdxw1l6xnl1opsJef6F+9DVnOqmdpvsPJl+j0KY+nW1NsG7bRMW14+1YwblVezFesXu0+2nMcQEtLP0I57RlbzOOJ2pjGQ1ZxXveRJ18bKy76Uj2xQK+fqaPKq5iqkn5EvFlRzNdVc7crzyHpGv3QH1Tx/c795//7dxpfpTVqylnP05SXZah433eWL7UNyDMPYPGyMRWBvwtHpthln6buzMONHouGAlK6itPOlB7EMoxnQq6l6Q+lj6qunax7+1u64+JvG9znRYqcK0G/v+qWjI3LF83Y77b/pdT2sZLohj3LfI1c2G+Ng5D2A9yYK196sVnXGx+zXrrJ2ZNqN+gfZCTd1abeZdqUyVvN2HoBhbFY2xgwg9K4hDk47dE/zZYhWNMtcoy+8nMgsPRwLJ4m9E2PjZOpmHieJdxPq6WxZj3Wnso2rOk1PI58O15XGE0qjorbypAvF6VGP/r2kR2NefHfBxXcP6XtRfvG1nxutk/bfK/+8qfslt93dzjfs7moYxqZhYyiAYO4Idvl4wrk0J3L5tYD0TGDWczTz1wMBH3snbgjDKYZivt88W2ZkU2U0nIc68+923HXf7T0IJpvY0Y0LgezlLI/1kDzp2JXmeMmQ1nV+/pftOtqU1c+a8Mu1IH4TllQSN55JJeTzg9aZxtlUidQwddDFEnJnA7s297atQT9DM9iy159DXAmHX7fO4dett2Wqhd/7xg97+aokHfKt7p3RrXniJ5q6v3X7ze19AOsb45+QYRgvnI3xv3d4cJtG/uwPLQIP5WsNjIMrZGomUSgXu03HJ4oudka8ezoLYRXVPbv14fZoWSs57oLn1LMFutAsAmumrR23ADpwI/UDey92bQrtWkkWgRebqJuSgfgs5Upbltqbm8ppv9M5g9rvH+gvOluXVEK+RrOprMrcpyUUlN91i8AsFS2vp+qoL2exIJ9t1jfm9hYnPhvBMIxNw8ZYA8Bv6gqunD6mj/vuXVy6tbNTtw5ZIfEGAskUyswFPAv7B3IlWyyogpum4ENP+JG6n2Gs7azciDsDreHY9d51VCGEUwBacf61cHF11i+qW7Z5twmsiZcDrjNXFLLGbOXcUJPrXBvFFBRWrshqjoZgceL2NtTTFYOLhhaeVZCl3Ad8U55/pS9XiIu/2k1iAeVKb8cK60en3AHzrb0I7r3XMxVb59Y45pXQjm+WLL3EtzX8RrYbzDA2JRtDAfhNXan7ZjB/xABwyzl0h0bWwf8x/FWgUOdRk4SF0MSvvXnWd8w+yNvUoZy1XTjFIJBvW6c65nbPuvDOYYaQFDHIkkPlJe40lqrZwBYV1kBc315L467qlYlmtE1fok3oC7+QK4MstlNz/x6UuBFLe3UsR0qXtziWU26lJbf43b3gzGH9la6bOK3nzWzKn3AWzGDHDs5HpfXU3yW6w6qPm3S8PRiGYWxsNoYCiBuwgsGdkRmATvlQECOLq42XTHwuiXgZXEw1GV1LlYQw8IuY/R2+bJ9eLnaT0BGpr2XquePr8fb56FWkxA1YTbwIF9JZSgG/RKG5U1bBC0pqrzCURtFkGncox8JriWaiupuIpULdq+M7KmfddZjVNIJLfN9ZVlNL5sJhdxr5q7nKv2/lxVcdYO9jlzWPh381Xg7DMDYnG2MNAFz/HezeaQiFMKouG1NK6+NH1BJG0TS2c/el3fkDbpbgnX7CBqfenqX4nAhcdPmxlk18nN07BrDbMnC7jUNTMl9HKkfoXP3u3NauXnw9ndpNTMR516ReTSLNe5HatTFbKJBC3SfUVXiTT5kh2/qjnX/yHqQSso7bTFYca0JTSEYT9kJhZdCN7+LtN36tvSmuO6yQDcPYLGwMBRBG2H5xNA1DEBdLg0kkLKCGT4jt08+aPOmO3WCD9zF1ZJC5AHLgn3UhqPv9wtW97sotP3dRzO/KYXSxOsi2ljc2exKvpKE4OVKJCwXtmyvJObxSCgx8+0PsnxAaWojvJf1ehzN/K4lt1rwxF3Uen4rlBe+itE5U2H2HW7yuC5oYS/2smS2VGQcfvjR2+p+698YhLyDbCGYYm5WNYQKS8bF8NO2rwuap49qbE9NM2v+HzVdZkqeW9ncSk4/vz469LIntMyTLcJoKrQPnx+YNs4mhdY54aHt4JrQvKLM00mlKWDQO+eLicXMYzfpl7Tb4Wlv1P/GT7sHjbejSXONBNy25x1wbhrG52BgzgBDNM5pGGjdQGe5gks1aTQC1oefS7JUw951i9EB17x8fD1CfG7TrHyPLSPnhsJpOe4+AqDcPDQe2851170BB74CTKbYvG6onqT/K6b9na1lcqI0je/949zlfbk2z0cxHBo0HvyeEPRKd55ORfKYxOJyUwvyjjevnz73mS613MfL7GIaxadgYM4AQzXM47r5Ky+0ypKWmGJXRtFb2XFm6phwJt6yCG0GHzVxLnaEHx9Q7TPD0GWSjWYe9jnTMyLyWppooX1vO0Ia0jbVfEB8JGwH0Ly6bGU46K+m13VIDpT8nYLA9edG1xBPOtFAWr2ve3//1lde1m2XB4Axj03LOZgAi8mMi8qiI7BWR953aQ9oeacvQKHo4z7jPcJFhdJyGfWjtJfAzgIW8Ve/s40V7pJuEPxhbR3oAfJjNpIfT+1F8uqM3hqAIciTyxKrHtTGM6rvV6EwlbdPRvCkv7LIeynf11QfGvtt0ZpLPDGL63N72e7ENYYaxeTknCkBEcuD3gLcC1wM/IyLXn/ihk3Qk4zr3U5Bei9q5Y6ZpvvNNO69yS3uqsXzV0BrAuDOBkzpaI/ah7zEtKSeuTRznTF0NbqRjb/q/63lyiE6z8c3lEcodzeY3LeoxLrTwxP6LXZYhb57oOVQL1UozO1p6cfu9jOyvMAxj03CuZgA3A3tV9TFV7QOfAG45bu6wQBrdY5Jb41w7Q75SGjfRtBNM+83gUpmYWOIB88l5tvlK5u75pN62tZPucI2D9o7GOETxXiJPrNfHLZLcjbDDge+NXE3b0oPXw7uJHjw+udi6HuMKBVnUl+HWJRoTVxpaIn03nZ7r0IupQfwdpBayqcq3rXYzCS9X70A7ZMTxTG+GYWx8ztUawBXA08n3Z4C/nWYQkduA2/zX9afe9c8ePEeybAQuBp4730KcQ6x9m5cLuW1w4bfvujN5+LwtAqvq7cDtACJyj6redL5kOddY+zY3F3L7LuS2wWS070yeP1cmoH3AnuT7bp9mGIZhbBDOlQL4G+BaEblKRLrArcCd56guwzAM4zQ4JyYgVS1F5D3AX+D21n5UVR86wSO3nws5NhDWvs3Nhdy+C7ltYO07IaJqbhyGYRiTyMYIBWEYhmF8zzEFYBiGMaGcdwVwWiEjNjgi8oSIfENE7g9uWiKyQ0Q+JyLf9n+3n285TwUR+aiIHBKRB5O0sW0Rxwf9b/mAiNx4/iQ/NY7Tvl8XkX3+97tfRN6W3PsV375HReQt50fqU0dE9ojIXSLysIg8JCLv9emb/jc8QdsuiN9PRKZE5G4R+bpv32/49KtE5Ku+HX/kHW0QkZ7/vtffv/KklajqefvgFoi/A1wNdIGvA9efT5nOUrueAC4eSvs3wPv89fuAf32+5TzFtvwwcCPw4MnaArwN+C+4fcKvAb56vuU/zfb9OvC/jMl7vf832gOu8v928/PdhpO0bxdwo7+eB77l27Hpf8MTtO2C+P38bzDnrzvAV/1v8kngVp/+74D/0V//Y+Df+etbgT86WR3newbwwkJGbG5uAe7w13cAbz9/opw6qvpXwJGh5OO15RbgY+r4CrBNRHZ9TwQ9TY7TvuNxC/AJVV1X1ceBvbh/wxsWVd2vqvf560XgEdxO/U3/G56gbcdjU/1+/jdY8l87/qPAG4A/8enDv134Tf8EeKOInDCezflWAONCRpzoB9wsKPCXInKvD3kBsFNV9/vrA8DO8yPaWeF4bbmQfs/3eBPIRxNz3aZunzcJvAo3krygfsOhtsEF8vuJSC4i9wOHgM/hZi1HVTVEZUzbENvn7x8DLjpR+edbAVyovE5Vb8RFQ323iPxwelPdHO2C8L+9kNqS8CHgGuAGYD/w2+dVmrOAiMwBfwr8oqoupPc2+284pm0XzO+nqpWq3oCLpnAz8NKzWf75VgAXZMgIVd3n/x4C/hz3wx0MU2n/99D5k/CMOV5bLojfU1UP+v94NfBhGjPBpmyfiHRwHeQfqOqf+eQL4jcc17YL7fcDUNWjwF3AD+LMcmETb9qG2D5/fytw+ETlnm8FcMGFjBCRWRGZD9fAm4EHce16p8/2TuDT50fCs8Lx2nIn8LPek+Q1wLHEzLBpGLJ5/wTu9wPXvlu9t8VVwLXA3d9r+V4I3gb8EeARVf2d5Nam/w2P17YL5fcTkUtEZJu/ngbehFvnuAv4KZ9t+LcLv+lPAV/0s7vjswFWut+GW73/DvCr51ues9Ceq3GeBl8HHgptwtnivgB8G/g8sON8y3qK7flD3DR6gLM3vut4bcF5Lfye/y2/Adx0vuU/zfb9Ry//A/4/1a4k/6/69j0KvPV8y38K7XsdzrzzAHC//7ztQvgNT9C2C+L3A14BfM2340HgX/r0q3GKay/wx0DPp0/573v9/atPVoeFgjAMw5hQzrcJyDAMwzhPmAIwDMOYUEwBGIZhTCimAAzDMCYUUwCGYRgTiikAwzCMCcUUgGEYxoTy/wO6Q/Ou8mOF4AAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# M1 (low index) changes only\n",
    "plt.pcolormesh(J_diff.cpu().detach().numpy())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "9d3a49fc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Hold signal M1 (low index) changes only\n",
    "plt.pcolormesh(S_diff.cpu().detach().numpy())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "24c5a6ca",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Bias M1 (low index) changes only\n",
    "plt.pcolormesh(B_diff.cpu().detach().numpy())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a022ccae",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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
}