Competition_Example_Notebook.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "0a38a23b",
   "metadata": {},
   "source": [
    "# Example Notebook: Neuron Synapse Competition"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "66105ff6",
   "metadata": {},
   "source": [
    "## Load in and Wrangle Training Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "e2237b6d",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "#load in training data on each potential synapse\n",
    "data = pd.read_csv(\"./train_data.csv\")\n",
    "\n",
    "#load in additional features for each neuron\n",
    "feature_weights = pd.read_csv(\"./feature_weights.csv\")\n",
    "morph_embeddings = pd.read_csv(\"./morph_embeddings.csv\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c4070b48",
   "metadata": {},
   "source": [
    "#### Basic info on these dataframes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "21377c2f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Main Dataframe Size: (185832, 30)\n",
      "\n",
      "Main Dataframe:\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 185832 entries, 0 to 185831\n",
      "Data columns (total 30 columns):\n",
      " #   Column                          Non-Null Count   Dtype  \n",
      "---  ------                          --------------   -----  \n",
      " 0   ID                              185832 non-null  int64  \n",
      " 1   axonal_coor_x                   185832 non-null  int64  \n",
      " 2   axonal_coor_y                   185832 non-null  int64  \n",
      " 3   axonal_coor_z                   185832 non-null  int64  \n",
      " 4   dendritic_coor_x                185832 non-null  int64  \n",
      " 5   dendritic_coor_y                185832 non-null  int64  \n",
      " 6   dendritic_coor_z                185832 non-null  int64  \n",
      " 7   adp_dist                        185832 non-null  float64\n",
      " 8   post_skeletal_distance_to_soma  185832 non-null  float64\n",
      " 9   pre_skeletal_distance_to_soma   185832 non-null  float64\n",
      " 10  pre_oracle                      185832 non-null  float64\n",
      " 11  pre_test_score                  185832 non-null  float64\n",
      " 12  pre_rf_x                        185832 non-null  float64\n",
      " 13  pre_rf_y                        185832 non-null  float64\n",
      " 14  post_oracle                     185832 non-null  float64\n",
      " 15  post_test_score                 185832 non-null  float64\n",
      " 16  post_rf_x                       185832 non-null  float64\n",
      " 17  post_rf_y                       185832 non-null  float64\n",
      " 18  compartment                     185832 non-null  object \n",
      " 19  pre_brain_area                  185832 non-null  object \n",
      " 20  post_brain_area                 185832 non-null  object \n",
      " 21  pre_nucleus_x                   185832 non-null  int64  \n",
      " 22  pre_nucleus_y                   185832 non-null  int64  \n",
      " 23  pre_nucleus_z                   185832 non-null  int64  \n",
      " 24  post_nucleus_x                  185832 non-null  int64  \n",
      " 25  post_nucleus_y                  185832 non-null  int64  \n",
      " 26  post_nucleus_z                  185832 non-null  int64  \n",
      " 27  pre_nucleus_id                  185832 non-null  int64  \n",
      " 28  post_nucleus_id                 185832 non-null  int64  \n",
      " 29  connected                       185832 non-null  bool   \n",
      "dtypes: bool(1), float64(11), int64(15), object(3)\n",
      "memory usage: 41.3+ MB\n",
      "\n",
      "Feature Weights Size: (2692, 513)\n",
      "\n",
      "Feature Weights:\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 2692 entries, 0 to 2691\n",
      "Columns: 513 entries, nucleus_id to feature_weight_511\n",
      "dtypes: float64(512), int64(1)\n",
      "memory usage: 10.5 MB\n",
      "\n",
      "Morphological Embeddings Size: (2667, 33)\n",
      "\n",
      "Morphological Embeddings:\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 2667 entries, 0 to 2666\n",
      "Columns: 33 entries, nucleus_id to morph_emb_31\n",
      "dtypes: float64(32), int64(1)\n",
      "memory usage: 687.7 KB\n"
     ]
    }
   ],
   "source": [
    "print(\"Main Dataframe Size:\",data.shape)\n",
    "print(\"\\nMain Dataframe:\")\n",
    "data.info()\n",
    "\n",
    "print(\"\\nFeature Weights Size:\",feature_weights.shape)\n",
    "print(\"\\nFeature Weights:\")\n",
    "feature_weights.info(verbose=False)\n",
    "\n",
    "print(\"\\nMorphological Embeddings Size:\",morph_embeddings.shape)\n",
    "print(\"\\nMorphological Embeddings:\")\n",
    "morph_embeddings.info(verbose=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9157b5f4",
   "metadata": {},
   "source": [
    "See the documentation for additional information on these features.  The main dataframe contains features for each potential synapse (each row).  The outcome to predict is 'connected'.  Note that many of the raw features in this dataframe will not be directly useful for prediction (e.g. x,y,z coordiantes, pre- and post-synapse neuron ID and etc.).  You will need to creatively engineer new features that will be more useful for prediction.\n",
    "\n",
    "The additional information on feature weights and morphological embeddings are given for each separate neuron (the neuron ID is matched with that in the main dataframe).  These will need to be merged with the other data to create training and test sets. "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "49737cbf",
   "metadata": {},
   "source": [
    "#### Merge Data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2b83752d",
   "metadata": {},
   "source": [
    "Before merging, we concatenate the feature weights and morphological embeddings into np.array's to decrease memory usage."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "2a78bac9",
   "metadata": {},
   "outputs": [],
   "source": [
    "# join all feature_weight_i columns into a single np.array column\n",
    "feature_weights[\"feature_weights\"] = (\n",
    "    feature_weights.filter(regex=\"feature_weight_\")\n",
    "    .sort_index(axis=1)\n",
    "    .apply(lambda x: np.array(x), axis=1)\n",
    ")\n",
    "# delete the feature_weight_i columns\n",
    "feature_weights.drop(\n",
    "    feature_weights.filter(regex=\"feature_weight_\").columns, axis=1, inplace=True\n",
    ")\n",
    "\n",
    "# join all morph_embed_i columns into a single np.array column\n",
    "morph_embeddings[\"morph_embeddings\"] = (\n",
    "    morph_embeddings.filter(regex=\"morph_emb_\")\n",
    "    .sort_index(axis=1)\n",
    "    .apply(lambda x: np.array(x), axis=1)\n",
    ")\n",
    "# delete the morph_embed_i columns\n",
    "morph_embeddings.drop(\n",
    "    morph_embeddings.filter(regex=\"morph_emb_\").columns, axis=1, inplace=True\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "3f87211b",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = (\n",
    "    data.merge(\n",
    "        feature_weights.rename(columns=lambda x: \"pre_\" + x), \n",
    "        how=\"left\", \n",
    "        validate=\"m:1\",\n",
    "        copy=False,\n",
    "    )\n",
    "    .merge(\n",
    "        feature_weights.rename(columns=lambda x: \"post_\" + x),\n",
    "        how=\"left\",\n",
    "        validate=\"m:1\",\n",
    "        copy=False,\n",
    "    )\n",
    "    .merge(\n",
    "        morph_embeddings.rename(columns=lambda x: \"pre_\" + x),\n",
    "        how=\"left\",\n",
    "        validate=\"m:1\",\n",
    "        copy=False,\n",
    "    )\n",
    "    .merge(\n",
    "        morph_embeddings.rename(columns=lambda x: \"post_\" + x),\n",
    "        how=\"left\",\n",
    "        validate=\"m:1\",\n",
    "        copy=False,\n",
    "    )\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "210ce030",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Data Size: (185832, 34)\n",
      "\n",
      "Data:\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 185832 entries, 0 to 185831\n",
      "Data columns (total 34 columns):\n",
      " #   Column                          Non-Null Count   Dtype  \n",
      "---  ------                          --------------   -----  \n",
      " 0   ID                              185832 non-null  int64  \n",
      " 1   axonal_coor_x                   185832 non-null  int64  \n",
      " 2   axonal_coor_y                   185832 non-null  int64  \n",
      " 3   axonal_coor_z                   185832 non-null  int64  \n",
      " 4   dendritic_coor_x                185832 non-null  int64  \n",
      " 5   dendritic_coor_y                185832 non-null  int64  \n",
      " 6   dendritic_coor_z                185832 non-null  int64  \n",
      " 7   adp_dist                        185832 non-null  float64\n",
      " 8   post_skeletal_distance_to_soma  185832 non-null  float64\n",
      " 9   pre_skeletal_distance_to_soma   185832 non-null  float64\n",
      " 10  pre_oracle                      185832 non-null  float64\n",
      " 11  pre_test_score                  185832 non-null  float64\n",
      " 12  pre_rf_x                        185832 non-null  float64\n",
      " 13  pre_rf_y                        185832 non-null  float64\n",
      " 14  post_oracle                     185832 non-null  float64\n",
      " 15  post_test_score                 185832 non-null  float64\n",
      " 16  post_rf_x                       185832 non-null  float64\n",
      " 17  post_rf_y                       185832 non-null  float64\n",
      " 18  compartment                     185832 non-null  object \n",
      " 19  pre_brain_area                  185832 non-null  object \n",
      " 20  post_brain_area                 185832 non-null  object \n",
      " 21  pre_nucleus_x                   185832 non-null  int64  \n",
      " 22  pre_nucleus_y                   185832 non-null  int64  \n",
      " 23  pre_nucleus_z                   185832 non-null  int64  \n",
      " 24  post_nucleus_x                  185832 non-null  int64  \n",
      " 25  post_nucleus_y                  185832 non-null  int64  \n",
      " 26  post_nucleus_z                  185832 non-null  int64  \n",
      " 27  pre_nucleus_id                  185832 non-null  int64  \n",
      " 28  post_nucleus_id                 185832 non-null  int64  \n",
      " 29  connected                       185832 non-null  bool   \n",
      " 30  pre_feature_weights             185832 non-null  object \n",
      " 31  post_feature_weights            185832 non-null  object \n",
      " 32  pre_morph_embeddings            138123 non-null  object \n",
      " 33  post_morph_embeddings           185832 non-null  object \n",
      "dtypes: bool(1), float64(11), int64(15), object(7)\n",
      "memory usage: 48.4+ MB\n"
     ]
    }
   ],
   "source": [
    "print(\"Data Size:\",data.shape)\n",
    "print(\"\\nData:\")\n",
    "data.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "63fc35ec",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(32,)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data[\"pre_morph_embeddings\"][0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "93e71263",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "basal           97842\n",
      "axon            40979\n",
      "oblique         15833\n",
      "apical          14952\n",
      "apical_tuft      7338\n",
      "apical_shaft     6937\n",
      "soma             1951\n",
      "Name: compartment, dtype: int64\n",
      "\n",
      "V1    74001\n",
      "RL    69168\n",
      "AL    42663\n",
      "Name: pre_brain_area, dtype: int64\n",
      "\n",
      "V1    78199\n",
      "RL    77505\n",
      "AL    30128\n",
      "Name: post_brain_area, dtype: int64\n"
     ]
    }
   ],
   "source": [
    "print(data[\"compartment\"].value_counts())\n",
    "print()\n",
    "print(data[\"pre_brain_area\"].value_counts())\n",
    "print()\n",
    "print(data[\"post_brain_area\"].value_counts())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "798015c5",
   "metadata": {},
   "source": [
    "## Basic Exploratory Analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "bcdb53dc",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "from tqdm import tqdm\n",
    "from sklearn.metrics.pairwise import cosine_similarity"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2dc227ee",
   "metadata": {},
   "source": [
    "### Unbalanced Data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "660421ac",
   "metadata": {},
   "source": [
    "The outcomes, the adps that form a connection or synapse ('connected'==True) and the adps that do not synapse ('connected'==False), are unbalanced:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "13ff436c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number Connected (Synapses): 1366\n",
      "\n",
      "Number Not Connected: 184466\n"
     ]
    }
   ],
   "source": [
    "print(\"Number Connected (Synapses):\",data['connected'].sum())\n",
    "\n",
    "print(\"\\nNumber Not Connected:\",(~data['connected']).sum())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1abf1022",
   "metadata": {},
   "source": [
    "The number of presynaptic neurons and postsynaptic neurons are also unbalanced:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "44a8917f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "All the adps are from 77 pre- neurons and 2663 post- neurons.\n"
     ]
    }
   ],
   "source": [
    "print(f\"All the adps are from {data['pre_nucleus_id'].nunique()} pre- neurons and {data['post_nucleus_id'].nunique()} post- neurons.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e8c02cc6",
   "metadata": {},
   "source": [
    "### Exploration: Synaptic Proximity Distance "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "a6163e8c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.FacetGrid at 0x1061b3880>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 500x500 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.catplot(data=data, x='connected', y='adp_dist', kind='boxen')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "80f50d61",
   "metadata": {},
   "source": [
    "Neurons that form a synpase ('connected') tend to be closer together (smaller distance between pre- and post-synaptic ADP - 'adp_dist')."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1d6717c6",
   "metadata": {},
   "source": [
    "## Example: Basic Feature Engineering"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5d306806",
   "metadata": {},
   "source": [
    "### Exploration: Neurons that form a synpase tend to have similar functional neural responses"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "97bc6105",
   "metadata": {},
   "source": [
    "Neurons in the visual cortex each respond differently to visual stimuli.  This is often captured by a neuron's tuning function, or the visual features that drive a neuron to fire.  The feature weight vectors ('feature_weights') are a kind of summary of each neuron's tuning function.  Neuroscience suggests that neurons with similar tuning are slightly more likely to be connected.  To assess this, we would need to engineer a new feature that capture the tuning similarity of the pre- and post- synaptic neurons for each ADP.  (See the documentation for more details).  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "e6679b5c",
   "metadata": {},
   "outputs": [],
   "source": [
    "#cosine similarity function\n",
    "def row_feature_similarity(row):\n",
    "    pre = row[\"pre_feature_weights\"]\n",
    "    post = row[\"post_feature_weights\"]\n",
    "    return (pre * post).sum() / (np.linalg.norm(pre) * np.linalg.norm(post))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "3388f91f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# compute the cosine similarity between the pre- and post- feature weights\n",
    "data[\"fw_similarity\"] = data.apply(row_feature_similarity, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "2bd80efe",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.FacetGrid at 0x1c98ee87460>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWAAAAFgCAYAAACFYaNMAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAAfL0lEQVR4nO3de5TcZZ3n8fe3q7vTuUIqN0IgF7ajEh1YkjYuqyjacQ6wBowTBxQvw4RFUbRnnRmNu7jOrKMgjJ7By2xAgjjKmZxBjCSYgbXSIs6gkk4kCUm41AkJBELS6Q65kHT69t0/6pdMXyrdVdVV9aSqPq9z6nTV8/tV/b6E6k//6qnneX7m7oiISPFVhS5ARKRSKYBFRAJRAIuIBKIAFhEJRAEsIhJIdegCCuGKK67wRx99NHQZIiInWbrGsjwDPnDgQOgSRESGVZYBLCJSChTAIiKBKIBFRAJRAIuIBKIAFhEJRAEsIhKIAlhEJBAFsIhIIApgEZFAFMAiIoEogEVEAlEAi4gEogCuUE1NTTQ1NYUuQ6SiKYBFRAIpy/WAZWhf//rX2bp1KxMmTAhdikhF0xlwBXryySfp7e3l8OHDoUsRqWgKYBGRQBTAIiKBKIBFRAJRAIuIBBI8gM3sCjN7zsySZrb8NPtcbmZPm9k2M/t1sWsUESmEoMPQzCwGfB94P7AH2GBma9x9e599zgb+EbjC3V8ys6lBihURybPQZ8ALgaS773T3TmAVcM2AfT4K/MzdXwJw9/1FrlFEpCBCB/AM4OU+j/dEbX29CZhoZo+b2UYz+0TRqhMRKaDQM+EsTZsPeFwNLAAagdHAb83sd+7+fL8XMrsJuAlg5syZBShVRCS/Qp8B7wHO7/P4PODVNPs86u5vuPsB4Ang4oEv5O73uHuDuzdMmTKlYAWLiORL6ADeAMw1szlmVgtcB6wZsM/DwGVmVm1mY4B3ADuKXKeISN4F7YJw924zuwV4DIgB97n7NjP7dLR9hbvvMLNHgS1AL3Cvuz8TrmoRkfwI3QeMu68D1g1oWzHg8Z3AncWsS0Sk0EJ3QYiIVCwFsIhIIArgCtTV1QWA+8ARfyJSTArgCtTd3Q0ogEVCUwCLiASiABYRCUQBLCISiAJYRCQQBXAF0pdvImcGBXAFUgCLnBkUwCIigSiARUQCUQCLyIi0traGLqFkKYBFJGdbtmzhwx/+MFu2bAldSklSAItITrq7u7ntttsAuP32209NcZfMKYBFJCerV6/m4MGDALS3t7N69erAFZUeBbCIZK2trY2VK1fS0dEBQEdHBytXrqS9vT1wZaVFASwiWWtubqa3t7dfW29vL+vXrw9UUWlSAItI1hobG6mq6h8fVVVVNDY2BqqoNCmARSRr8XicZcuWUVdXB0BdXR3Lli0jHo8Hrqy0KIBFJCdLliw5FbjxeJwlS5YErqj0KIBFJCfV1dUsX74cgOXLl1NdHfwi6yVHASwiEogCWERyookYI6cAFpGcaCLGyCmARSRrmoiRHwpgEclac3MzPT09/do0ESN7CmARyVpjY+OgK6v09vZqIkaWFMAikjV316Wt8kABLCJZa25uHjQV2czUBZElBbCIZK2xsZFYLNavLRaLqQsiSwpgEclaPB5nwYIF/doaGhq0FkSWFMAikrW2tjZaWlr6tW3YsEHD0LKkABaRrDU3Nw+a+dbd3a0+4CwpgEUka/Pnzx80Drinp2dQt4QMTQEsIlnbtGlT2gXZN27cGKii0qQAFpGsNTY2UlNT06+tpqZGoyCypAAWkazF43FuvPHGU2sAV1dXc+ONN2oURJYUwCKSk6uvvvrUbDh35+qrrw5cUelRAItITtasWXNqMkYsFmPNmjWBKyo9CmARyVpbWxt33303nZ2dAHR2dmo5yhwogEUka1qOMj8UwCKStXSjHaqqqjQKIksKYBHJWjweZ/r06ace19XVsWzZMo2CyJICWERyMnny5FP34/E4S5YsCVhNaVIAi0hOzOzU/eXLl58aEyyZUwCLyIjMmzePiy66KHQZJUkBXGHWrl075GORbA2ckiyZUwBXmEQiQay2jvHnzCZWW0cikQhdkpSgtWvXkkwmAUgmk/pDniMFcAUaEz+HC6+6gTHxc0KXIiUqkUhAbw9/9KbZ0NujP+Q5UgCLSE4uOP8c7vjiMi44X3/Ic6UAFhEJRAEsIhKIAlhEJBAFsIhIIApgEZFAFMAiIoEED2Azu8LMnjOzpJktH2K/t5tZj5ktLWZ9IiKFEjSAzSwGfB+4EpgHfMTM5p1mv28CjxW3QhGRwgl9BrwQSLr7TnfvBFYB16TZ73PAQ8D+YhYnIlJIoQN4BvByn8d7orZTzGwGsARYMdQLmdlNZtZiZi2tra15L1REJN9CB7ClafMBj/8B+JK796TZ9z+e5H6Puze4e8OUKVPyVV/ZWLt2LU1NTSSTSY617WXHuh9yrG0vyWSSpqYmLaYiEkDoFZT3AOf3eXwe8OqAfRqAVdHiz5OBq8ys291/XpQKy0QikWDr9mcZNXEatWOhu7eX2onT6AW2bn8WgMWLF4ctUqTChA7gDcBcM5sDvAJcB3y07w7uPufkfTO7H3hE4ZubUROnMavxY4Pad6//SYBqpBStXbuWRCKRWoqyt5sv3rGSnS/vhapWmpqaWLRokf6QZyFoALt7t5ndQmp0Qwy4z923mdmno+1D9vuKSHElEgmSz+1gzrmTAPDuE8yZnroQZ/K5HYA+SWUj9Bkw7r4OWDegLW3wuvufFaMmETm9OedO4rZPfWBQ+5fvfiRANaUt9JdwIiIVSwEsIhKIAlhEJBAFsIhIIApgEZFAFMAiIoGY+8CZv6WvoaHBW1paQpcR1MkB8yclk0mOd3YzauK0QfueOLiP0bXV1NfXn2rTgHqB9O8j7+nigmgccF87X23DYjV6H6WXbtmF8OOApTASiQRbtj1LzVlTUw11carroKdn8B/c6glT6QJ27GkHoOtQatE5/eJIIpHghWe3MWfqeABmx2uBWryrY9C+c6aMBaC7/SUAXtx/BND7aCgK4DJWc9ZUplx2XdbPa/3NqgJUI6VqztTx/N31C7N+3q0PPFWAasqL+oBFRAJRAIuIBKIAFhEJRAEsIhKIAlhEJBAFsIhIIApgEZFAFMAiIoEogEVEAtFMOJEKN3C9h76SySTefSKnWW0v7j+CtSdpampKu13rRCiARSpeIpHghR1bmT2pbtC2WRMAavHuE1m/bmrdCKdr/wuDtu1qS60loQCWkjXcmUvXie6c1nXoOrSfZEe7zlwqyOxJdfzt4llFO95X1+4u2rHOZArgEpZIJNi8bQex8VMGb6w9m6pa6Ontzfp1q8ZPpgN45qUDg7b1HGkFdOYikg8K4BIXGz+FsQv/pGjHe+Oph4p2LJFyp1EQIiKBKIBFRAJRAIuIBKIAFhEJRAEsIhJIRgFsZn9vZm8tdDEiIpUk0zPgZ4F7zOz3ZvZpMzurkEWJiFSCjMYBu/u9wL1m9mbgBmCLmf078AN3/1UhCxSR4Q01K3I4yWQS7+oo6uy0XW0d2OHTrxMxnHKZjZnxRAwziwFviW4HgM3AF8zsU+6e/bXPRSRvEokEL2zfzKyzsv9aZ+ZoYDR4V0f+CzuN1BoTx+l8ZWvWz919KDW7s2IC2My+DSwGmoFvuPvJpZG+aWbPFao4EcncrLOq+Mrlo0OXUXBfe/x46BLyJtMz4GeAW939WJptC/NYj4hIxcj088r1A8PXzNYDuPuhvFclIlIBhjwDNrM6YAww2cwmAhZtmgCcW+DaRETK2nBdEJ8C/oJU2G7q034Y+H6BaqooI/32uudEV1FXKOs50koy+XrFf3stkg9DBrC73wXcZWafc/fvFqmmipJIJNj8zA4YNyn7J1dPgOrc1vzN2dhJvAFs3rU/++cebQPK49trkXwYrgvife7eDLxiZh8auN3df1awyirJuElw8TWhqyi8zQ+HrkDkjDJcF8R7SA09S3fK4oACWEQkR8N1QXzVzKqAf3X3fylSTSIiFWHYYWju3gvcUoRaREQqSqbjgH9pZn9lZuebWfzkraCViYiUuUxnwv159POzfdocuCC/5YiIVI5MV0ObU+hCREQqTTarob0NmAfUnWxz938qRFEiIpUg09XQvgpcTiqA1wFXAv8GKIBFRHKU6ZdwS4FG4DV3vwG4GBhVsKpERCpApgF8PBqO1m1mE4D96As4EZERybQPuMXMzgZ+AGwEjgJPDfkMkQo1kgWWcpVMJvETvWW1WPnp7H69Fzue++WMclWIhaQyHQXxmejuCjN7FJjg7lvyWolImUgkEryw7Q/MHNdTtGOeXwPUgHcX7ZDBzBwH8DondrcU7ZgvHY0B+V9IarjFeOYPtc3dN51uu0glmzmuh/85/3DoMiRPvrFpQkFed7gz4G8Nsc2B9+WxFhGRijLcYjzvLVYhZ4JQfXd0dFXGUo1H20gmD5dF351IPmS0HnC6tYCh/NYDTiQSPP3MDnrGFHGZi6rxqYs+9XrxjhnKmDiHgI079xXtkLFj7YAWgZczk9YDHqBnTJzjb7kqdBmSJ6OfXRe6BJHTGnY94OjnDcUpR0SkcmQ6Ffls4BPA7L7PcffPj7QAM7sCuAuIAfe6++0Dtl8PfCl6eBS42d03j/S4IiKhZToRYx3wO2ArkLcrQJpZjNTVld8P7AE2mNkad9/eZ7cXgfe4+0EzuxK4B3hHvmoQEQkl0wCuc/cvFOD4C4Gku+8EMLNVwDXAqQB29yf77P874LwC1CEiUnSZrgXxYzP772Y2Pc9XxJgBvNzn8Z6o7XSWAf+aboOZ3WRmLWbW0tramofSREQKK9Mz4E7gTuB/kRr9APm5IoalaUs7HsvM3ksqgN+Vbru730Oqe4KGhoYKGNMlIqUu0wD+AlDv7gfyfPw9wPl9Hp8HvDpwJzO7CLgXuNLd2/Jcg4hIEJl2QWwDjhXg+BuAuWY2x8xqgeuANX13MLOZpMYbf9zdny9ADSIiQWR6BtwDPG1mvwJOnGwc6TA0d+82s1uAx0gNQ7vP3beZ2aej7SuA/w1MAv7RzAC63b1hJMcVETkTZBrAP49ueefu60gNc+vbtqLP/RuBGwtxbBGRkDJdD/hHhS5ERKTSDLcYz7+4+5+a2Vb6j04wwN39ooJWJyJSxoY7Az65buAHCl2IiEilGXIUhLvvje4eAF52992kroZ8MWmGi4mISOYyHYb2BFBnZjOA9cANwP2FKkpEpBJkGsDm7seADwHfdfclwLzClSUiUv4yDmAzuxS4HvhF1JbpEDYREUkj0wBuAr4MrI4mSlwA/KpwZYmIlL9MxwE/Qaof+OTjncCpWXBm9l13/1z+yxMRKV/56kZ4Z55eR6Qs7Dt8gsd39YQuQ/Jk3+ETzCzA62baBSEiInmmL9JECmDahFFcPvvE8DtKSXiyfVRBXjdfAZxuYfWS1HHkIF2vbB9+RykJduQgMC10GSJpZdQFEY16GMpdeahFRKSiZHoGfH80C24DqdEQv3H3rSc3uvv9BagtiLrxE/EZmmNSLuqO7ApdgshpZToM7d3RFSveDlwO/MLMxrl7Pi7MKSJSkTIKYDN7F3BZdDsbeAT4TeHKEhEpf5l2QfwaaAFuA9a5e2fhShIRqQyZBvAkUpMt3g183sx6gd+6+1cKVpmISJnLtA/4dTPbSeoS8ucB/xWoKWRhIiLlbshhaGb24+hnL/AtIA6sAN7s7u8pfHkiIuVruDPgBWY2C3gG+Fif9nFmhru3F640EZHyNlwArwAeBeaQ+hLuJCN1kc7hJmiIiMhpDHdNuO+4+4XAD939gj63Oe6u8BURGYGMpiK7+82FLkREpNJoOUoRkUAUwCIigSiARUQCUQCLiASiABYRCUQBLCISiK4JJ1IALx2N8Y1NE0KXIXny0tEYcwvwugpgkTxbtGhR0Y+ZTCbxE28w6+zy/1C7+/VebNRY6uvri3bMuRTm/6sCWCTPFi9ezOLFi4t6zKamJjpf2cpXLh9d1OOG8LXHj1M7o5677ir9S1EqgAeIHWtn9LPrQpcheRI71o6uiixnKgVwH6E+Or7R0QXjJhX92EV3tI2xdTVF/egI04L8fxXJhAK4j1AfHTfv2g8XX1PU4wax+WHqZ08ti4+OIvlQ/j32IiJnKAWwiEggCmARkUAUwCIigSiARUQCUQCLiASiABYRCUQBLCISiAJYRCQQBbCISCCaiixSJnYf6uVrjx8PXUbB7T7Uy9wZoavIDwWwSBkYyYJDyWQS7zrO7El1eaxoaLvaOrCa0TktzDR3RpiFswpBASxSBkaykFRTUxNd+1/gbxfPynNVp/fVtbupmVoea/qOhAL4THC0DTY/HLqKwjvaBkwNXYXIGUMBHNhIPzq+caKL2PgpeaxoaD1HWhk7Ktc1faeWzUdHkXxQAAc20o+Oz7x0gLEL/yTPVZ3eG089RP3MyRX/0VEkHzQMTUQkEAWwiEggCmARkUCCB7CZXWFmz5lZ0syWp9luZvadaPsWM5sfok4RkXwLGsBmFgO+D1wJzAM+YmbzBux2JTA3ut0E/N+iFikiUiChz4AXAkl33+nuncAqYODlga8B/slTfgecbWbTi12oiEi+hQ7gGcDLfR7vidqy3Qczu8nMWsyspbW1Ne+FiojkW+hxwJamzXPYB3e/B7gHoKGhYdB2ETm9XW0dfHXt7qIeb64mRQYP4D3A+X0enwe8msM+IpKjoWYnJpNJvPsEc6aOz/p1X9x/BKselXbW5Nyp5bOgzkiEDuANwFwzmwO8AlwHfHTAPmuAW8xsFfAO4JC77y1umWeuniOtvPHUQ0U9Hkwu2vGk8IaajdnU1ER3+0v83fULs37dWx94iur4TM2aHELQAHb3bjO7BXgMiAH3ufs2M/t0tH0FsA64CkgCx4AbQtV7phnuzOXYiW5qzsr+c17Xof2MGVV9mvUeJuvMRSRPQp8B4+7rSIVs37YVfe478Nli11UKhjtz2bGnnSmXXZf167b+ZhX158V15iJSYKFHQYiIVCwFsIhIIApgEZFAFMAiIoEogEVEAgk+CkJEzmwv7j/CrQ88ldPz5sYLUFAZUQCLyGkNHPOdTCbxni4uOHfSoH13vtqGxf7jeoFz45rtNhwFcBnrOrSf1t+syul5nKdTFxk81rypqYmeI/u57VMfGLTvl+9+hNj4qRo/ngUFcJlKd+ZyvLObUROnDdr3xMF9jK7tM/PtvLjOXESKQAFcptKduTy/9yCzGj82aN/d639C/fSJOnMRKTKNghARCUQBLCISiAJYRCQQBbCISCAKYBGRQBTAIiKBaBhaBTlxcB+71/8kbTvTJwaoSKSyKYArxMmJFclkko7ObsZMms6xtr3U1VbzR/PeookXkrEXX23jy3c/kra9/s261HE2FMAV4uTEjKamJnbuO8SFV93AjnU/5IJpZ2kChmSs7x9yeru54Pzp7Hx5L1RVU//mC/WHPEsKYBHJWN8/5L3HD3HHF5fxxTtWUjVaf8hzoS/hREQCUQCLiASiABaRnOx8+TW+eMdKdr78WuhSSpYCuAIda3+NHet+yLF2/eJIbhYtWgRVMbY+vwuqYvryLUcK4AqzaNEiejo7OPLaLno6O/SLIzlZvHjxqfWj6+vr+y19KplTAFeYgb8o+sURCUcBLCIj0tXVFbqEkqUAFpER2b59O1u2bAldRklSAItITtz91P3bb7+d7u7ugNWUJgWwiOTkwIED/e6vXr06YDWlSQEsIllra2tj7969px53dnZy77330t7eHrCq0qMAFpGsNTc3D2rr6upi/fr1AaopXQpgEcna/Pnz+/UBA/T29rJgwYJAFZUmBbCIZG3Tpk2YWb82M2Pjxo2BKipNCmARyVq6M2B31xlwlhTAIpK1TZs2UVXVPz6qqqp0BpwlBbCIZK2xsZHq6v7Xc6iurqaxsTFQRaVJASwiWYvH47z97W/v17Zw4ULi8XigikqTAlhEstbW1jaou6GlpUXjgLOkABaRrDU3N9PT09OvraenR+OAs6QAFpGsNTY20tvb26/N3dUHnCUFsIhkbeAQtNO1ydAUwCKStebm5rTjgNUFkR0FsIhkbf78+YO6IDQVOXsKYBHJ2qZNm4jFYv3aYrGYJmJkSQEsIllrbGykpqamX1tNTY2+hMuSAlhEshaPx1m2bBm1tbUA1NbWsmzZMk3EyJICuAINXMVKJBdLlixh8uTJAEyePJklS5YErqj0KIArkAJY8qG6uprly5cDsHz58kFrQ8jw9C8mIjm76KKLePDBB5kyZUroUkqSzoBFZEQUvrlTAIuIBKIArkAn++rUFywSlgK4Ap0cv6kAFglLASwiEogCWEQkEAWwiEggwQLYzOJm9kszeyH6OTHNPueb2a/MbIeZbTOzphC1lpuJE1P/1Bo4LxJWyDPg5cB6d58LrI8eD9QN/KW7Xwj8F+CzZjaviDWWpWuvvZaxY8cyderU0KWIVLSQAXwN8KPo/o+ADw7cwd33uvum6P4RYAcwo1gFlqvFixdTX1/PpEmTQpciUtFCBvA0d98LqaAFhjwdM7PZwCXA70+z/SYzazGzltbW1nzXWpYOHz4cugSRilbQADazhJk9k+Z2TZavMw54CPgLd0+bGu5+j7s3uHuDpkYO75VXXuHFF1/ke9/7XuhSRCpWQQPY3Re5+9vS3B4G9pnZdIDo5/50r2FmNaTC9wF3/1kh660Ux44d48CBAwD89Kc/5dixY4ErEqlMIbsg1gCfjO5/Enh44A6Wmqq1Etjh7t8uYm1l7eabb+73+DOf+UygSkQqW8gAvh14v5m9ALw/eoyZnWtm66J93gl8HHifmT0d3a4KU2552LhxI7t37+7XtmvXLv7whz8EqkikctnAS0uXg4aGBm9paQldxhnp2muvZd++fYPazznnHFatWhWgIpGKkHbhFc2EExEJRAFcYU5eQmagL33pS0WuREQUwBXmkksuYebMmf3aZs6cySWXXBKoIpHKpQCuQCtWrBjysYgUhwK4Ao0ZM4alS5cCsHTpUsaMGRO4IpHKpOWwKtQtt9zCggULuPTSS0OXIlKxdAZcwRS+ImEpgEVEAlEAi4gEogAWEQlEASwiEogCWEQkEAWwiEggCmARkUDKcjlKM2sFdg+7o0wGDoQuQkqe3kfDO+DuVwxsLMsAlsyYWYu7N4SuQ0qb3ke5UxeEiEggCmARkUAUwJXtntAFSFnQ+yhH6gMWEQlEZ8AiIoEogEVEAtGC7GXEzHqArX2aPujuu06z71F3H1eUwqTkmNkkYH308BygB2iNHi90984ghZUZ9QGXkWxCVQEsmTKzvwGOuvvf92mrdvfucFWVB3VBlDEzG2dm681sk5ltNbNr0uwz3cyeMLOnzewZM7ssav9jM/tt9NwHzUxhXeHM7H4z+7aZ/Qr4ppn9jZn9VZ/tz5jZ7Oj+x8zsqeh9dbeZxULVfSZTAJeX0dEb/mkzWw10AEvcfT7wXuBbZmYDnvNR4DF3/8/AxcDTZjYZuBVYFD23BfhC0f4r5Ez2JlLvi7883Q5mdiFwLfDO6H3VA1xfnPJKi/qAy8vx6A0PgJnVAN8ws3cDvcAMYBrwWp/nbADui/b9ubs/bWbvAeYB/x7ldS3w2+L8J8gZ7kF37xlmn0ZgAbAhev+MBvYXurBSpAAub9cDU4AF7t5lZruAur47uPsTUUD/N+DHZnYncBD4pbt/pNgFyxnvjT73u+n/Kfrke8uAH7n7l4tWVYlSF0R5OwvYH4Xve4FZA3cws1nRPj8AVgLzgd8B7zSz+mifMWb2piLWLaVhF6n3C2Y2H5gTta8HlprZ1GhbPHqfyQA6Ay5vDwBrzawFeBp4Ns0+lwN/bWZdwFHgE+7eamZ/BvyzmY2K9rsVeL7gFUspeQj4hJk9Taor63kAd99uZrcC/8/MqoAu4LNoidhBNAxNRCQQdUGIiASiABYRCUQBLCISiAJYRCQQBbCISCAKYJERMrPZZvbRHJ53v5ktLURNUhoUwCIjN5vUmhoiWVEAS8kys0+Y2RYz22xmPzazWdHqb1uinzOj/e43s++Y2ZNmtvPkWaeZXW5mj5vZT83sWTN74ORiRWa2wMx+bWYbzewxM5setdebWSI65iYz+0/A7cBl0SJI/8PMYmZ2p5ltiGr5VPRcM7Pvmdl2M/sFMDXIP5ycOdxdN91K7ga8FXgOmBw9jgNrgU9Gj/+c1OJCAPcDD5I64ZgHJKP2y4FDwHnRtt8C7wJqgCeBKdF+1wL3Rfd/T2qFOUitfTAmep1H+tR2E3BrdH8UqdXk5gAfAn4JxIBzgdeBpaH/LXULd9NUZClV7wN+6u4HANy93cwuJRVyAD8G7uiz/8/dvRfYbmbT+rQ/5e57AKIptbNJBePbgF9GJ8QxYK+ZjQdmuPvq6Jgd0fMG1vbHwEV9+nfPAuYC7wb+2VOrib1qZs0j+QeQ0qcAllJlwHDz6PtuPzHguenae0j9Thiwzd0v7XdAswlZ1PY5d39swPOvyqBmqSDqA5ZStR740+jaZZhZnFS3wXXR9uuBf8vxtZ8DpkRn1JhZjZm91d0PA3vM7INR+ygzGwMcAcb3ef5jwM3RGsuY2ZvMbCzwBHBd1Ec8ndQi+VLBdAYsJcndt5nZ14FfW+pipH8APk9qcfm/JnUByRtyfO3OqPvgO2Z2Fqnfk38AtgEfB+42s/9DapWvDwNbgG4z20yqv/kuUl0Zm6Iv9VqBDwKrSXWdbCW1ctivc6lPyodWQxMRCURdECIigSiARUQCUQCLiASiABYRCUQBLCISiAJYRCQQBbCISCD/Hw7Rvp0pLljjAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#visualize these results\n",
    "sns.catplot(data=data, x='connected', y='fw_similarity', kind='boxen')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "01035751",
   "metadata": {},
   "source": [
    "There may be a very, very slight difference, but we can investigate this further."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "106744b9",
   "metadata": {},
   "source": [
    "### Exploration: Similar functional nerual responses by projection region in the brain"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7fe3de1b",
   "metadata": {},
   "source": [
    "Projection regions indicate the brain regions where the pre-synaptic and post-synaptic neurons are located in the brain.  Each brain region is responsible for different neural functions.  We can engineer a new feature to capture the projection regions and perhaps see if similar neural tuning is affiliated with connecitons in particular projection regions.  For example, it has been shown that connected neurons have slightly more similar neural tuning than unconnected neurons in the V1->V1 projection region."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "29908dfb",
   "metadata": {},
   "outputs": [],
   "source": [
    "# generate projection group as pre->post\n",
    "data[\"projection_group\"] = (\n",
    "    data[\"pre_brain_area\"].astype(str)\n",
    "    + \"->\"\n",
    "    + data[\"post_brain_area\"].astype(str)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "e7b83b7b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.FacetGrid at 0x1c9963f99d0>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 637.25x576 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# check the distribution of the cosine similarity for projections groups with more than 100 synapses\n",
    "projection_group_counts = data.query('connected')['projection_group'].value_counts()\n",
    "projection_group_counts = projection_group_counts[projection_group_counts > 100].index\n",
    "sns.displot(\n",
    "    data=data.query('projection_group in @projection_group_counts'), \n",
    "    x='fw_similarity', \n",
    "    hue='connected', \n",
    "    row='projection_group',\n",
    "    common_norm=False, \n",
    "    stat='probability',\n",
    "    binwidth=.02,\n",
    "    height=2,\n",
    "    aspect=4,\n",
    "    facet_kws={'sharey': False}\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "1582585b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 584.875x500 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.catplot(data=data, x='projection_group', y='fw_similarity', hue='connected', kind='bar', alpha=.7)\n",
    "plt.tight_layout()\n",
    "plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.);\n",
    "plt.xticks(rotation=90);"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2dc827ab",
   "metadata": {},
   "source": [
    "We have shown some simple examples of how to engineer new features and also shown that these are related to neuron connectivity.  You will likley need to explore and engineer many other possible features in this competition.  "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3a78d7f6",
   "metadata": {},
   "source": [
    "## Example: Simple Logistic Regression"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "937655df",
   "metadata": {},
   "source": [
    "Now that we have shown some features are associated with neuron connectivity, we fit a simple calssification model: Logistic Regression."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "54850f42",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy: 0.9927085855732236\n"
     ]
    }
   ],
   "source": [
    "# logistic regression model (connected ~ fw_similarity + adp)\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.metrics import balanced_accuracy_score, accuracy_score, confusion_matrix\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.pipeline import Pipeline\n",
    "\n",
    "# split into a train and test set \n",
    "#(Even though we're working with the competition training set, you may want to have your own internal train and test sets.)\n",
    "train_data, test_data = train_test_split(data, test_size=0.2, random_state=1)\n",
    "\n",
    "# create pipeline\n",
    "pipe = Pipeline(\n",
    "    [(\"scaler\", StandardScaler()), (\"model\", LogisticRegression(random_state=2))]\n",
    ")\n",
    "\n",
    "# fit model\n",
    "pipe.fit(train_data[[\"fw_similarity\", \"adp_dist\"]], train_data[\"connected\"])\n",
    "\n",
    "# predict on test data\n",
    "test_data[\"pred\"] = pipe.predict_proba(test_data[[\"fw_similarity\", \"adp_dist\"]])[:, 1]\n",
    "\n",
    "# compute accuracy\n",
    "print(f\"accuracy: {accuracy_score(test_data['connected'], test_data['pred'] > .5)}\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ec2397d5",
   "metadata": {},
   "source": [
    "Wow.  This looks like terrific accuracy.  But, remember the data is unbalanced.  Let's check the confusion matrix."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "2b351fdd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[36896,     0],\n",
       "       [  271,     0]])"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "confusion_matrix(test_data['connected'], test_data['pred'] > .5)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e4faf555",
   "metadata": {},
   "source": [
    "The model always predicts not connected!  We need a new evaluation metric."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5157b7fb",
   "metadata": {},
   "source": [
    "### Balanced Accuracy Metric"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b83390f1",
   "metadata": {},
   "source": [
    "Sensitivity = True Positive Rate = True Positive / (True Positive + False Negative)\n",
    "\n",
    "Specificity = True Negative Rate = True Negative / (True Negative + False Positive)\n",
    "\n",
    "Balanced Accuracy = (Sensitivity + Specificity)/2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "2481e7b8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "balanced accuracy: 0.5\n"
     ]
    }
   ],
   "source": [
    "print(\n",
    "    f\"balanced accuracy: {balanced_accuracy_score(test_data['connected'], test_data['pred'] > .5)}\"\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5947b122",
   "metadata": {},
   "source": [
    "## Data Augmentation for Unbalanced Classification"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5b4b933c",
   "metadata": {},
   "source": [
    "As our simple model always predicts not connected, we need to use some technique that balances the two classes.  While there are many strategies one can employ, we demostrate a simple random over sampling strategy that uses the bootstrap to augment the data for the minority ('connected') class.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "b45c9e7e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy: 0.647752038098313\n",
      "[[23859 13037]\n",
      " [   55   216]]\n",
      "balanced accuracy: 0.7218517172433216\n"
     ]
    }
   ],
   "source": [
    "from imblearn.over_sampling import RandomOverSampler\n",
    "\n",
    "# oversample connected neuron pairs\n",
    "ros = RandomOverSampler(random_state=0)\n",
    "X_resampled, y_resampled = ros.fit_resample(\n",
    "    train_data[[\"fw_similarity\", \"adp_dist\"]], train_data[\"connected\"]\n",
    ")\n",
    "\n",
    "# fit model\n",
    "pipe.fit(X_resampled, y_resampled)\n",
    "\n",
    "# predict on test data\n",
    "test_data[\"pred\"] = pipe.predict_proba(test_data[[\"fw_similarity\", \"adp_dist\"]])[:, 1]\n",
    "\n",
    "# compute accuracy\n",
    "print(f\"accuracy: {accuracy_score(test_data['connected'], test_data['pred'] > .5)}\")\n",
    "\n",
    "# confusion matrix\n",
    "print(confusion_matrix(test_data['connected'], test_data['pred'] > .5))\n",
    "\n",
    "# compute balanced accuracy\n",
    "print(\n",
    "    f\"balanced accuracy: {balanced_accuracy_score(test_data['connected'], test_data['pred'] > .5)}\"\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "48d1a156",
   "metadata": {},
   "source": [
    "Simple data augmentation improves our balanced accuracy dramatically!"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bb82ab85",
   "metadata": {},
   "source": [
    "# Create Example Prediction File for Leaderboard Submission"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0d12af71",
   "metadata": {},
   "source": [
    "### Load and Merge Leaderboard Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "a8bdc8d8",
   "metadata": {},
   "outputs": [],
   "source": [
    "#we need to first load and merge the leaderboard data to have the same format as the training set\n",
    "lb_data = pd.read_csv(\"./leaderboard_data.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "110c67f3",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "basal           23008\n",
      "axon             9146\n",
      "oblique          3658\n",
      "apical           3099\n",
      "apical_shaft     1606\n",
      "apical_tuft      1595\n",
      "soma              481\n",
      "Name: compartment, dtype: int64\n",
      "\n",
      "AL    17489\n",
      "V1    13076\n",
      "RL    12028\n",
      "Name: pre_brain_area, dtype: int64\n",
      "\n",
      "V1    19177\n",
      "RL    16159\n",
      "AL     7257\n",
      "Name: post_brain_area, dtype: int64\n"
     ]
    }
   ],
   "source": [
    "print(lb_data[\"compartment\"].value_counts())\n",
    "print()\n",
    "print(lb_data[\"pre_brain_area\"].value_counts())\n",
    "print()\n",
    "print(lb_data[\"post_brain_area\"].value_counts())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "4d0b97fd",
   "metadata": {},
   "outputs": [],
   "source": [
    "lb_data = (\n",
    "    lb_data.merge(\n",
    "        feature_weights.rename(columns=lambda x: \"pre_\" + x), \n",
    "        how=\"left\", \n",
    "        validate=\"m:1\",\n",
    "        copy=False,\n",
    "    )\n",
    "    .merge(\n",
    "        feature_weights.rename(columns=lambda x: \"post_\" + x),\n",
    "        how=\"left\",\n",
    "        validate=\"m:1\",\n",
    "        copy=False,\n",
    "    )\n",
    "    .merge(\n",
    "        morph_embeddings.rename(columns=lambda x: \"pre_\" + x),\n",
    "        how=\"left\",\n",
    "        validate=\"m:1\",\n",
    "        copy=False,\n",
    "    )\n",
    "    .merge(\n",
    "        morph_embeddings.rename(columns=lambda x: \"post_\" + x),\n",
    "        how=\"left\",\n",
    "        validate=\"m:1\",\n",
    "        copy=False,\n",
    "    )\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "4074a460",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<bound method NDFrame.head of           ID  axonal_coor_x  axonal_coor_y  axonal_coor_z  dendritic_coor_x  \\\n",
       "0          0         527499         544761         912282            529457   \n",
       "1          1        1269890         471870         837816           1271230   \n",
       "2          2         666057         469875         925239            665387   \n",
       "3          3         696633         453516         925806            696648   \n",
       "4          4         567525         585921         888426            570714   \n",
       "...      ...            ...            ...            ...               ...   \n",
       "42588  42588         765513         818622         870114            766346   \n",
       "42589  42589         784476         609000         828177            782564   \n",
       "42590  42590         766038         773808         861693            763428   \n",
       "42591  42591         738276         614460         855099            741617   \n",
       "42592  42592         769314         457443         874419            771959   \n",
       "\n",
       "       dendritic_coor_y  dendritic_coor_z  adp_dist  \\\n",
       "0                543459            916958   4783.37   \n",
       "1                469651            841045   4060.72   \n",
       "2                467214            923430   2912.80   \n",
       "3                453417            921875   3272.83   \n",
       "4                583676            891264   4441.75   \n",
       "...                 ...               ...       ...   \n",
       "42588            817181            871011   1597.63   \n",
       "42589            606911            828562   2607.11   \n",
       "42590            773490            862976   2648.36   \n",
       "42591            612387            854028   3843.24   \n",
       "42592            456564            869856   4978.29   \n",
       "\n",
       "       post_skeletal_distance_to_soma  pre_skeletal_distance_to_soma  ...  \\\n",
       "0                           239163.00                       178032.0  ...   \n",
       "1                           294573.00                      1193910.0  ...   \n",
       "2                             4102.75                       387714.0  ...   \n",
       "3                            31123.00                       424714.0  ...   \n",
       "4                           114201.00                       250411.0  ...   \n",
       "...                               ...                            ...  ...   \n",
       "42588                        74187.70                       426027.0  ...   \n",
       "42589                       172794.00                       553175.0  ...   \n",
       "42590                        44174.50                       372663.0  ...   \n",
       "42591                       178179.00                       517147.0  ...   \n",
       "42592                       321244.00                       763630.0  ...   \n",
       "\n",
       "       pre_nucleus_z  post_nucleus_x  post_nucleus_y  post_nucleus_z  \\\n",
       "0             861240          566528          777344          920800   \n",
       "1             861240         1196352          492736          919560   \n",
       "2             861240          669760          470336          913720   \n",
       "3             861240          669760          470336          913720   \n",
       "4             861240          672384          609728          932720   \n",
       "...              ...             ...             ...             ...   \n",
       "42588         837120          807168          757376          869080   \n",
       "42589         837120          807168          757376          869080   \n",
       "42590         837120          807168          757376          869080   \n",
       "42591         837120          807168          757376          869080   \n",
       "42592         837120          807168          757376          869080   \n",
       "\n",
       "       pre_nucleus_id  post_nucleus_id  \\\n",
       "0              582471           199477   \n",
       "1              582471           518848   \n",
       "2              582471           256716   \n",
       "3              582471           256716   \n",
       "4              582471           260802   \n",
       "...               ...              ...   \n",
       "42588          294484           335443   \n",
       "42589          294484           335443   \n",
       "42590          294484           335443   \n",
       "42591          294484           335443   \n",
       "42592          294484           335443   \n",
       "\n",
       "                                     pre_feature_weights  \\\n",
       "0      [0.19683373, 0.19128098, 0.49339047, -0.029781...   \n",
       "1      [0.19683373, 0.19128098, 0.49339047, -0.029781...   \n",
       "2      [0.19683373, 0.19128098, 0.49339047, -0.029781...   \n",
       "3      [0.19683373, 0.19128098, 0.49339047, -0.029781...   \n",
       "4      [0.19683373, 0.19128098, 0.49339047, -0.029781...   \n",
       "...                                                  ...   \n",
       "42588  [0.4097931, 0.2948376, 0.05925691, 0.111899965...   \n",
       "42589  [0.4097931, 0.2948376, 0.05925691, 0.111899965...   \n",
       "42590  [0.4097931, 0.2948376, 0.05925691, 0.111899965...   \n",
       "42591  [0.4097931, 0.2948376, 0.05925691, 0.111899965...   \n",
       "42592  [0.4097931, 0.2948376, 0.05925691, 0.111899965...   \n",
       "\n",
       "                                    post_feature_weights  \\\n",
       "0      [-0.03441501, 0.41964892, 0.23311342, 0.327173...   \n",
       "1      [-0.03917461, -0.1830603, -0.3704222, 0.183293...   \n",
       "2      [-0.66653997, 0.5036552, 0.047438737, 0.181578...   \n",
       "3      [-0.66653997, 0.5036552, 0.047438737, 0.181578...   \n",
       "4      [-0.40308818, 0.20905656, 0.25083357, -0.19731...   \n",
       "...                                                  ...   \n",
       "42588  [-0.2947989, -0.35604966, 0.7178404, 0.1323682...   \n",
       "42589  [-0.2947989, -0.35604966, 0.7178404, 0.1323682...   \n",
       "42590  [-0.2947989, -0.35604966, 0.7178404, 0.1323682...   \n",
       "42591  [-0.2947989, -0.35604966, 0.7178404, 0.1323682...   \n",
       "42592  [-0.2947989, -0.35604966, 0.7178404, 0.1323682...   \n",
       "\n",
       "                                    pre_morph_embeddings  \\\n",
       "0      [0.5603107213973999, -0.3908874094486236, -0.1...   \n",
       "1      [0.5603107213973999, -0.3908874094486236, -0.1...   \n",
       "2      [0.5603107213973999, -0.3908874094486236, -0.1...   \n",
       "3      [0.5603107213973999, -0.3908874094486236, -0.1...   \n",
       "4      [0.5603107213973999, -0.3908874094486236, -0.1...   \n",
       "...                                                  ...   \n",
       "42588                                                NaN   \n",
       "42589                                                NaN   \n",
       "42590                                                NaN   \n",
       "42591                                                NaN   \n",
       "42592                                                NaN   \n",
       "\n",
       "                                   post_morph_embeddings  \n",
       "0      [1.132591724395752, -0.0064769592136144, -0.67...  \n",
       "1      [1.0723994970321655, -0.7540942430496216, 0.11...  \n",
       "2      [-0.0201839506626129, -0.2563340067863464, 0.1...  \n",
       "3      [-0.0201839506626129, -0.2563340067863464, 0.1...  \n",
       "4      [-0.0254557952284812, 0.7573301792144775, -0.2...  \n",
       "...                                                  ...  \n",
       "42588  [1.4674314260482788, 0.6272068619728088, 0.410...  \n",
       "42589  [1.4674314260482788, 0.6272068619728088, 0.410...  \n",
       "42590  [1.4674314260482788, 0.6272068619728088, 0.410...  \n",
       "42591  [1.4674314260482788, 0.6272068619728088, 0.410...  \n",
       "42592  [1.4674314260482788, 0.6272068619728088, 0.410...  \n",
       "\n",
       "[42593 rows x 33 columns]>"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lb_data.shape\n",
    "lb_data.head"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b6d054a",
   "metadata": {},
   "source": [
    "### Example Prediction: Simple Logistic Regression"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cc2467bc",
   "metadata": {},
   "source": [
    "Now we fit a our logistic regression model on the full training data and use it to make predictions on the leaderboard data. But first, we need to add in our new engineered feature on neural tuning similarity.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "089e47f4",
   "metadata": {},
   "outputs": [],
   "source": [
    "# compute the cosine similarity between the pre- and post- feature weights\n",
    "lb_data[\"fw_similarity\"] = lb_data.apply(row_feature_similarity, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "7a8e702c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# oversample connected neuron pairs for full training data\n",
    "ros = RandomOverSampler(random_state=0)\n",
    "X_resampled, y_resampled = ros.fit_resample(\n",
    "    data[[\"fw_similarity\", \"adp_dist\"]], data[\"connected\"]\n",
    ")\n",
    "\n",
    "# fit model\n",
    "pipe.fit(X_resampled, y_resampled)\n",
    "\n",
    "# predict on leaderboard data\n",
    "lb_data[\"pred\"] = pipe.predict_proba(lb_data[[\"fw_similarity\", \"adp_dist\"]])[:, 1]\n",
    "\n",
    "#create a boolean prediction solution\n",
    "lb_data[\"connected\"] = lb_data[\"pred\"] > .5"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "69c96a3f",
   "metadata": {},
   "source": [
    "### Example of Creating a Leaderboard Submission File"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "771bb7d2",
   "metadata": {},
   "outputs": [],
   "source": [
    "#columns should be ID, connected\n",
    "submission_data = lb_data.filter(['ID','connected'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "9ad284b2",
   "metadata": {},
   "outputs": [],
   "source": [
    "#writing csv files\n",
    "submission_data.to_csv('example_submission_data.csv',index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3a64e906",
   "metadata": {},
   "source": [
    "# Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "95b9b8f7",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "     class 0       0.33      1.00      0.50         1\n",
      "     class 1       1.00      0.50      0.67         4\n",
      "\n",
      "    accuracy                           0.60         5\n",
      "   macro avg       0.67      0.75      0.58         5\n",
      "weighted avg       0.87      0.60      0.63         5\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import classification_report\n",
    "y_true = [0, 1, 1, 1, 1]\n",
    "y_pred = [0, 0, 1, 0, 1]\n",
    "target_names = ['class 0', 'class 1']\n",
    "a=classification_report(y_true, y_pred, target_names=target_names)\n",
    "print(a)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "8761a61e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "str"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(a)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "ac10ad75",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from sklearn.metrics import precision_recall_fscore_support\n",
    "y_true = np.array([1,0,1,1,0])\n",
    "y_pred = np.array([1,1,0,0,0])\n",
    "precision,recall,fscore,support=precision_recall_fscore_support(y_true, y_pred, labels=[0,1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "f16d5994",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.33333333, 0.5       ])"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "precision"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "26a0ee4c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.5       , 0.33333333])"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "recall"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f3c43ee3",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}