tutorials/example_single_cell.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "OMP: Info #276: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead.\n"
     ]
    }
   ],
   "source": [
    "from spectrae import Spectra, SpectraDataset\n",
    "from torchdrug import datasets\n",
    "import numpy as np\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "from tqdm import tqdm\n",
    "# To run this tutorial please pip install cell-gears and torch_geometric if you do not already have these packages installed  \n",
    "from gears.pertdata import PertData\n",
    "from gears.gears import GEARS\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SPECTRA Single Cell Perturbation Tutorial\n",
    "\n",
    "### We take an example norman single-cell perturbation dataset from GEARS, taking the mean logfold change for each perturbation\n",
    "### We define a continuous spectral property based on a l2 norm between two perturbation expression vectors. We return the negative of this value since the algorithm expects a larger value means more similar.\n",
    "### We define cross-split overlap as average l2 norm between perturbation expression vectors in the test versus the train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "class SC_Perturb_Dataset(SpectraDataset):\n",
    "    \n",
    "    def parse(self, input_file):\n",
    "        pert_data = PertData('./data')\n",
    "        pert_data.load(data_name = input_file)\n",
    "        self.adata = pert_data.adata\n",
    "        self.control_expression = self.adata[self.adata.obs['condition'] == 'ctrl'].X.toarray().mean(axis=0)\n",
    "        return [p for p in self.adata.obs['condition'].unique() if p != 'ctrl']\n",
    "\n",
    "    def __len__(self):\n",
    "        return len(self.samples)\n",
    "\n",
    "    def get_mean_logfold_change(self, perturbation):\n",
    "        perturbation_expression = self.adata[self.adata.obs['condition'] == perturbation].X.toarray().mean(axis=0)\n",
    "        logfold_change = np.nan_to_num(np.log2(perturbation_expression + 1) - np.log2(self.control_expression + 1))\n",
    "        return logfold_change\n",
    "    \n",
    "    def sample_to_index(self, sample):\n",
    "        if not hasattr(self, 'index_to_sequence'):\n",
    "            print(\"Generating index to sequence\")\n",
    "            self.index_to_sequence = {}\n",
    "            for i in tqdm(range(len(self))):\n",
    "                x = self.__getitem__(i)\n",
    "                self.index_to_sequence['-'.join(list(x))] = i\n",
    "\n",
    "        return self.index_to_sequence[sample]            \n",
    "\n",
    "    def __getitem__(self, idx):\n",
    "        perturbation = self.samples[idx]\n",
    "        return self.get_mean_logfold_change(perturbation)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "class SPECTRA_SC_Perturb(Spectra):\n",
    "\n",
    "    def spectra_properties(self, sample_one, sample_two):\n",
    "        return -np.linalg.norm(sample_one - sample_two)\n",
    "\n",
    "    def cross_split_overlap(self, train, test):\n",
    "        average_similarity = []\n",
    "\n",
    "        for i in test:\n",
    "            for j in train:\n",
    "                average_similarity.append(self.spectra_properties(i, j))\n",
    "\n",
    "        return np.mean(average_similarity)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Found local copy...\n",
      "Found local copy...\n",
      "Found local copy...\n",
      "These perturbations are not in the GO graph and their perturbation can thus not be predicted\n",
      "['RHOXF2BB+ctrl' 'LYL1+IER5L' 'ctrl+IER5L' 'KIAA1804+ctrl' 'IER5L+ctrl'\n",
      " 'RHOXF2BB+ZBTB25' 'RHOXF2BB+SET']\n",
      "Local copy of pyg dataset is detected. Loading...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "SC_Perturb_Dataset = SC_Perturb_Dataset('norman', 'norman')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "sc_spectra = SPECTRA_SC_Perturb(SC_Perturb_Dataset, binary = False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "File sc_norman_precalculated_spectra_properties already exists, set force_recalculate to True to recalculate\n"
     ]
    }
   ],
   "source": [
    "sc_spectra.pre_calculate_spectra_properties('sc_norman')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 37950/37950 [00:00<00:00, 363134.50it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Stats for SPECTRA property graph (SPG)\n",
      "Number of nodes: 276\n",
      "Number of edges: 37950\n",
      "Number of connected components: 1\n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generating SPECTRA split for spectral parameter 0.0 and dataset norman\n",
      "Number of samples in independent set: 276\n",
      "Train size: 220\tTest size: 56\n",
      "Cross split overlap: -4.8159260749816895\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.0 and dataset norman\n",
      "Number of samples in independent set: 276\n",
      "Train size: 220\tTest size: 56\n",
      "Cross split overlap: -4.60251522064209\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.0 and dataset norman\n",
      "Number of samples in independent set: 276\n",
      "Train size: 220\tTest size: 56\n",
      "Cross split overlap: -4.512292861938477\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.05 and dataset norman\n",
      "Number of samples in independent set: 226\n",
      "Train size: 180\tTest size: 46\n",
      "Cross split overlap: -4.9465131759643555\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.05 and dataset norman\n",
      "Number of samples in independent set: 233\n",
      "Train size: 186\tTest size: 47\n",
      "Cross split overlap: -4.982058048248291\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.05 and dataset norman\n",
      "Number of samples in independent set: 228\n",
      "Train size: 182\tTest size: 46\n",
      "Cross split overlap: -4.945944786071777\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.1 and dataset norman\n",
      "Number of samples in independent set: 181\n",
      "Train size: 144\tTest size: 37\n",
      "Cross split overlap: -5.278563022613525\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.1 and dataset norman\n",
      "Number of samples in independent set: 185\n",
      "Train size: 148\tTest size: 37\n",
      "Cross split overlap: -5.177643299102783\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.1 and dataset norman\n",
      "Number of samples in independent set: 184\n",
      "Train size: 147\tTest size: 37\n",
      "Cross split overlap: -5.101210594177246\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.15 and dataset norman\n",
      "Number of samples in independent set: 142\n",
      "Train size: 113\tTest size: 29\n",
      "Cross split overlap: -5.494086265563965\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.15 and dataset norman\n",
      "Number of samples in independent set: 150\n",
      "Train size: 120\tTest size: 30\n",
      "Cross split overlap: -5.253544330596924\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.15 and dataset norman\n",
      "Number of samples in independent set: 144\n",
      "Train size: 115\tTest size: 29\n",
      "Cross split overlap: -5.609643459320068\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.2 and dataset norman\n",
      "Number of samples in independent set: 114\n",
      "Train size: 91\tTest size: 23\n",
      "Cross split overlap: -5.5548787117004395\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.2 and dataset norman\n",
      "Number of samples in independent set: 117\n",
      "Train size: 93\tTest size: 24\n",
      "Cross split overlap: -5.532756805419922\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.2 and dataset norman\n",
      "Number of samples in independent set: 120\n",
      "Train size: 96\tTest size: 24\n",
      "Cross split overlap: -5.58986234664917\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.25 and dataset norman\n",
      "Number of samples in independent set: 99\n",
      "Train size: 79\tTest size: 20\n",
      "Cross split overlap: -5.735699653625488\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.25 and dataset norman\n",
      "Number of samples in independent set: 97\n",
      "Train size: 77\tTest size: 20\n",
      "Cross split overlap: -5.951452255249023\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.25 and dataset norman\n",
      "Number of samples in independent set: 99\n",
      "Train size: 79\tTest size: 20\n",
      "Cross split overlap: -5.8724188804626465\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.3 and dataset norman\n",
      "Number of samples in independent set: 86\n",
      "Train size: 68\tTest size: 18\n",
      "Cross split overlap: -5.71300745010376\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.3 and dataset norman\n",
      "Number of samples in independent set: 75\n",
      "Train size: 60\tTest size: 15\n",
      "Cross split overlap: -5.966411590576172\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.3 and dataset norman\n",
      "Number of samples in independent set: 78\n",
      "Train size: 62\tTest size: 16\n",
      "Cross split overlap: -6.343864917755127\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.35 and dataset norman\n",
      "Number of samples in independent set: 70\n",
      "Train size: 56\tTest size: 14\n",
      "Cross split overlap: -6.241590976715088\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.35 and dataset norman\n",
      "Number of samples in independent set: 65\n",
      "Train size: 52\tTest size: 13\n",
      "Cross split overlap: -6.025228977203369\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.35 and dataset norman\n",
      "Number of samples in independent set: 68\n",
      "Train size: 54\tTest size: 14\n",
      "Cross split overlap: -6.022736072540283\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.4 and dataset norman\n",
      "Number of samples in independent set: 56\n",
      "Train size: 44\tTest size: 12\n",
      "Cross split overlap: -6.616042613983154\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.4 and dataset norman\n",
      "Number of samples in independent set: 50\n",
      "Train size: 40\tTest size: 10\n",
      "Cross split overlap: -7.0314741134643555\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.4 and dataset norman\n",
      "Number of samples in independent set: 55\n",
      "Train size: 44\tTest size: 11\n",
      "Cross split overlap: -6.15494441986084\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.45 and dataset norman\n",
      "Number of samples in independent set: 47\n",
      "Train size: 37\tTest size: 10\n",
      "Cross split overlap: -6.750374794006348\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.45 and dataset norman\n",
      "Number of samples in independent set: 45\n",
      "Train size: 36\tTest size: 9\n",
      "Cross split overlap: -6.381750106811523\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.45 and dataset norman\n",
      "Number of samples in independent set: 43\n",
      "Train size: 34\tTest size: 9\n",
      "Cross split overlap: -6.229277610778809\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.5 and dataset norman\n",
      "Number of samples in independent set: 43\n",
      "Train size: 34\tTest size: 9\n",
      "Cross split overlap: -6.623985290527344\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.5 and dataset norman\n",
      "Number of samples in independent set: 38\n",
      "Train size: 30\tTest size: 8\n",
      "Cross split overlap: -6.730915546417236\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.5 and dataset norman\n",
      "Number of samples in independent set: 33\n",
      "Train size: 26\tTest size: 7\n",
      "Cross split overlap: -6.337135314941406\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.55 and dataset norman\n",
      "Number of samples in independent set: 31\n",
      "Train size: 24\tTest size: 7\n",
      "Cross split overlap: -6.180523872375488\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.55 and dataset norman\n",
      "Number of samples in independent set: 30\n",
      "Train size: 24\tTest size: 6\n",
      "Cross split overlap: -6.477444171905518\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.55 and dataset norman\n",
      "Number of samples in independent set: 34\n",
      "Train size: 27\tTest size: 7\n",
      "Cross split overlap: -6.719655513763428\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.6 and dataset norman\n",
      "Number of samples in independent set: 26\n",
      "Train size: 20\tTest size: 6\n",
      "Cross split overlap: -6.700960159301758\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.6 and dataset norman\n",
      "Number of samples in independent set: 25\n",
      "Train size: 20\tTest size: 5\n",
      "Cross split overlap: -6.644288539886475\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.6 and dataset norman\n",
      "Number of samples in independent set: 24\n",
      "Train size: 19\tTest size: 5\n",
      "Cross split overlap: -6.492367267608643\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.65 and dataset norman\n",
      "Number of samples in independent set: 23\n",
      "Train size: 18\tTest size: 5\n",
      "Cross split overlap: -6.624109745025635\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.65 and dataset norman\n",
      "Number of samples in independent set: 19\n",
      "Train size: 15\tTest size: 4\n",
      "Cross split overlap: -6.797684669494629\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.65 and dataset norman\n",
      "Number of samples in independent set: 22\n",
      "Train size: 17\tTest size: 5\n",
      "Cross split overlap: -6.917760372161865\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.7 and dataset norman\n",
      "Number of samples in independent set: 18\n",
      "Train size: 14\tTest size: 4\n",
      "Cross split overlap: -6.626152515411377\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.7 and dataset norman\n",
      "Number of samples in independent set: 17\n",
      "Train size: 13\tTest size: 4\n",
      "Cross split overlap: -7.594491004943848\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.7 and dataset norman\n",
      "Number of samples in independent set: 20\n",
      "Train size: 16\tTest size: 4\n",
      "Cross split overlap: -6.973229885101318\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.75 and dataset norman\n",
      "Number of samples in independent set: 13\n",
      "Train size: 10\tTest size: 3\n",
      "Cross split overlap: -6.321884632110596\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.75 and dataset norman\n",
      "Number of samples in independent set: 12\n",
      "Train size: 9\tTest size: 3\n",
      "Cross split overlap: -7.978428363800049\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.75 and dataset norman\n",
      "Number of samples in independent set: 13\n",
      "Train size: 10\tTest size: 3\n",
      "Cross split overlap: -6.92338752746582\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.8 and dataset norman\n",
      "Number of samples in independent set: 9\n",
      "Train size: 7\tTest size: 2\n",
      "Cross split overlap: -7.36692476272583\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.8 and dataset norman\n",
      "Number of samples in independent set: 10\n",
      "Train size: 8\tTest size: 2\n",
      "Cross split overlap: -7.850466251373291\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.8 and dataset norman\n",
      "Number of samples in independent set: 10\n",
      "Train size: 8\tTest size: 2\n",
      "Cross split overlap: -6.930935859680176\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.85 and dataset norman\n",
      "Number of samples in independent set: 7\n",
      "Train size: 5\tTest size: 2\n",
      "Cross split overlap: -7.872118949890137\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.85 and dataset norman\n",
      "Number of samples in independent set: 8\n",
      "Train size: 6\tTest size: 2\n",
      "Cross split overlap: -6.830763339996338\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.85 and dataset norman\n",
      "Number of samples in independent set: 11\n",
      "Train size: 8\tTest size: 3\n",
      "Cross split overlap: -7.870787143707275\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.9 and dataset norman\n",
      "Number of samples in independent set: 6\n",
      "Train size: 4\tTest size: 2\n",
      "Cross split overlap: -8.546489715576172\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.9 and dataset norman\n",
      "Split for SP_0.90_1 could not be generated since independent set only has one sample\n",
      "Generating SPECTRA split for spectral parameter 0.9 and dataset norman\n",
      "Number of samples in independent set: 6\n",
      "Train size: 4\tTest size: 2\n",
      "Cross split overlap: -7.771939277648926\n",
      "\n",
      "\n",
      "\n",
      "Generating SPECTRA split for spectral parameter 0.95 and dataset norman\n",
      "Split for SP_0.95_0 could not be generated since independent set only has one sample\n",
      "Generating SPECTRA split for spectral parameter 0.95 and dataset norman\n",
      "Split for SP_0.95_1 could not be generated since independent set only has one sample\n",
      "Generating SPECTRA split for spectral parameter 0.95 and dataset norman\n",
      "Split for SP_0.95_2 could not be generated since independent set only has one sample\n",
      "Generating SPECTRA split for spectral parameter 1.0 and dataset norman\n",
      "Split for SP_1.00_0 could not be generated since independent set only has one sample\n",
      "Generating SPECTRA split for spectral parameter 1.0 and dataset norman\n",
      "Split for SP_1.00_1 could not be generated since independent set only has one sample\n",
      "Generating SPECTRA split for spectral parameter 1.0 and dataset norman\n",
      "Split for SP_1.00_2 could not be generated since independent set only has one sample\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "spectra_parameters = {'number_repeats': 3, \n",
    "                      'random_seed': [42, 44, 46],\n",
    "                      'spectral_parameters': [\"{:.2f}\".format(i) for i in np.arange(0, 1.05, 0.05)],\n",
    "                      'force_reconstruct': True,\n",
    "                                              }\n",
    "\n",
    "sc_spectra.generate_spectra_splits(**spectra_parameters)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x175caa7c0>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "stats = sc_spectra.return_all_split_stats()\n",
    "plt.scatter(stats['SPECTRA_parameter'], stats['cross_split_overlap'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x175f66550>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter(stats['SPECTRA_parameter'], stats['train_size'])\n",
    "plt.scatter(stats['SPECTRA_parameter'], stats['test_size']) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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": "3.9.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}