format_polar_bears.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "86ce8245-b923-42ba-b1cf-b0a8bdb804d4",
   "metadata": {},
   "source": [
    "# Notebook to format data from \"Metabolic Rate, Body Composition, Foraging Success, Behavior, and GPS Locations of Female Polar Bears (Ursus maritimus), Beaufort Sea, Spring, 2014-2016 and Resting Energetics of an Adult Female Polar Bear: U.S. Geological Survey data release\", Pagano (2018).\n",
    "\n",
    "Data can be found [here](https://alaska.usgs.gov/products/data.php?dataid=174).\n",
    "\n",
    "See also [this paper](https://www.int-res.com/articles/esr2017/32/n032p019.pdf).\n",
    "\n",
    "Some awesome [video footage](https://www.youtube.com/watch?v=qXQBjZLNYT0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "77461baa-2d89-4ddf-b8b9-283f19b3313e",
   "metadata": {},
   "source": [
    "## Setup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "a9cd11e9-258b-4304-a21d-f826f276b73b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Outputs:\n",
    "# formatted dataset directory\n",
    "# dataset_metadata (yaml), includes\n",
    "## sample_rate\n",
    "## dataset_label_names (list)\n",
    "## dataset_feature_names (list)\n",
    "## clip_ids (list)\n",
    "# clip_data (directory), includes\n",
    "## data_files (multiple files)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "7470e077-e978-495a-b349-27cc5a6b2a40",
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "ec388900-f0ab-4e41-a56b-9597e4f4911f",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import h5py\n",
    "from matplotlib import pyplot as plt\n",
    "import glob\n",
    "import yaml\n",
    "import scipy.signal as signal\n",
    "import datetime\n",
    "import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "6bc728a2-a91e-4da2-abea-4b57ceb4e818",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Specify data filepaths\n",
    "\n",
    "raw_data_dir = '/home/jupyter/behavior_data_local/data/raw/pagano_bears'\n",
    "formatted_data_dir = '/home/jupyter/behavior_data_local/data/formatted/pagano_bears'\n",
    "\n",
    "if not os.path.exists(formatted_data_dir):\n",
    "    os.makedirs(formatted_data_dir)\n",
    "    \n",
    "features_fps = os.path.join(raw_data_dir, 'PolarBear_archival_logger_data_southernBeaufortSea_2014_2016_revised.csv')\n",
    "annotations_fps = os.path.join(raw_data_dir, 'PolarBear_video-derived_behaviors_southernBeaufortSea_2014_2016_revised_w_ACC_corrected_times.csv')\n",
    "\n",
    "clip_data_dir = os.path.join(formatted_data_dir, 'clip_data')\n",
    "if not os.path.exists(clip_data_dir):\n",
    "    os.makedirs(clip_data_dir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "98f244e7-ac03-48e2-868a-1b7f7dd15fea",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Dataset metadata\n",
    "dataset_metadata = {}\n",
    "sr = 16\n",
    "dataset_metadata['sr'] = sr ## from dataset documentation\n",
    "dataset_metadata['dataset_name'] = 'pagano_bears'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "3b1891a7-1ff4-4df0-9c00-ecf634c3870a",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = pd.read_csv(features_fps)\n",
    "annotations = pd.read_csv(annotations_fps)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "7fa104a6-5093-484f-978e-1cb0d65243ee",
   "metadata": {},
   "outputs": [],
   "source": [
    "individuals = sorted(data['Bear'].unique())\n",
    "individuals = [int(x) for x in individuals]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "ee92f4bf-e469-400e-839c-1939da873bae",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[3, 5, 6, 7, 8, 9]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "individuals"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "32181469-4db3-473b-aeb9-4e572327c138",
   "metadata": {},
   "source": [
    "We are going to drop bear 3 from analysis. This bear was recorded in 2014, and the annotations are only at 30 second resolution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "6e02d3e2-5da6-4351-8774-27f3b4c5bf51",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[5, 6, 7, 8, 9]"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = data[data['Bear'] != 3]\n",
    "individuals.remove(3)\n",
    "individuals"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "8952a2d7-61c1-4c4e-b37a-221745bb82f0",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Split tracks into 3 hour clips, so that life is more manageable\n",
    "# Keep track of which individual is associated to which clip_id\n",
    "\n",
    "data['clip_id'] = 'clip_' + str(0)\n",
    "\n",
    "clip_size_samples = 1728000 # This is 3 hours of recording at 16 Hz\n",
    "current_clip_id = 0\n",
    "clip_ids = []\n",
    "\n",
    "clip_id_to_individual_id = {}\n",
    "for individual in individuals:\n",
    "    indices = data[data['Bear'] == individual].index\n",
    "    x = 0\n",
    "    while x< len(indices):\n",
    "        selected = indices[x : x +clip_size_samples]\n",
    "        data.loc[selected, 'clip_id'] = 'clip_' + str(current_clip_id)\n",
    "        clip_ids.append('clip_' + str(current_clip_id))\n",
    "        clip_id_to_individual_id['clip_' + str(current_clip_id)] = individual\n",
    "        current_clip_id += 1\n",
    "        x += clip_size_samples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "7ea20959-639e-4f2c-8efa-945516302110",
   "metadata": {},
   "outputs": [],
   "source": [
    "## Integer codes for individuals\n",
    "\n",
    "dataset_metadata['individual_ids'] = individuals # There are 5 individuals with tag data\n",
    "dataset_metadata['clip_id_to_individual_id'] = clip_id_to_individual_id\n",
    "dataset_metadata['clip_ids'] = clip_ids"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "b719e122-9b2d-4fb1-a519-36f58a545c0c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{0: [5], 1: [6], 2: [7], 3: [8], 4: [9]}\n",
      "There are 5 individuals\n"
     ]
    }
   ],
   "source": [
    "# Set up data splits: divide individuals across folds for cross validation\n",
    "\n",
    "n_folds = 5\n",
    "dataset_metadata['n_folds'] = n_folds\n",
    "individuals_per_fold = {i : [] for i in range(n_folds)}\n",
    "\n",
    "for pointer, individual in enumerate(individuals):\n",
    "    individuals_per_fold[pointer % n_folds].append(individual)\n",
    "    \n",
    "dataset_metadata['individuals_per_fold'] = individuals_per_fold\n",
    "    \n",
    "print(individuals_per_fold)\n",
    "print(\"There are %d individuals\" % len(individuals))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "ed6d55ef-3574-4052-9506-6c2d9af412cf",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set up clip ids \n",
    "\n",
    "clip_ids_per_fold = {i : [] for i in range(n_folds)}\n",
    "for clip_id in clip_ids:\n",
    "    individual = clip_id_to_individual_id[clip_id]\n",
    "    for fold in range(n_folds):\n",
    "        if individual in individuals_per_fold[fold]:\n",
    "            clip_ids_per_fold[fold].append(clip_id)\n",
    "for fold in range(n_folds):\n",
    "    clip_ids_per_fold[fold] = sorted(clip_ids_per_fold[fold])\n",
    "\n",
    "dataset_metadata['clip_ids_per_fold'] = clip_ids_per_fold"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3fca914b-f4d0-4265-b18c-71cd7aca2dc5",
   "metadata": {},
   "source": [
    "## Create dataset-level metadata"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "b9343516-5a19-4882-9e32-1ef628a49667",
   "metadata": {},
   "outputs": [],
   "source": [
    "# for individual in individuals:\n",
    "#     print(annotations[annotations['Bear'] == individual]['Behavior'].value_counts())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "7ac34de9-3853-44e0-ae8c-2b7bed24f72e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Favor labels which are long enough in duration to last for the annotation period, and which are abundant\n",
    "\n",
    "obs_beh_to_beh_name = {\n",
    "    'laying' : 'rest',\n",
    "    'walking' : 'walk',\n",
    "    'mixed' : 'unknown',\n",
    "    'adjusting' : 'unknown', \n",
    "    'sitting or standing' : 'rest',\n",
    "    'head shake' : 'head_shake',\n",
    "    'sitting or laying' : 'rest',\n",
    "    'rolling in snow' : 'roll',\n",
    "    'unknown' : 'unknown',\n",
    "    'eating' : 'eat',\n",
    "    'digging' : 'dig',\n",
    "    'grooming' : 'groom',\n",
    "    'interacting' : 'unknown',\n",
    "    'smelling snow' : 'unknown',\n",
    "    'swimming' : 'swim',\n",
    "    'running' : 'run',\n",
    "    'eating snow' : 'eat',\n",
    "    'pouncing' : 'pounce',\n",
    "    'standing' : 'rest',\n",
    "    'running and walking' : 'unknown',\n",
    "    'carrying carcass' : 'unknown',\n",
    "    'breeding' : 'unknown',\n",
    "    'rubbing' : 'unknown',\n",
    "    'jumping' : 'unknown',\n",
    "    'breaking into liar' : 'unknown',\n",
    "    'sticking head out of water' : 'unknown',\n",
    "    'moving ice' : 'unknown',\n",
    "    'catches seal' : 'unknown',\n",
    "    'biting seal' : 'unknown',\n",
    "    'crawling through ice cavern' : 'unknown',\n",
    "    'sticking head into water' : 'unknown',\n",
    "    'hauling out of water' : 'unknown',\n",
    "    'standing up' : 'unknown',\n",
    "    'standing ' : 'rest',\n",
    "    'standing or sitting' : 'rest',\n",
    "    '#NAME?' : 'unknown'\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "444a3dc3-dfb0-4907-833e-8caf3eb4d88d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['unknown', 'dig', 'eat', 'groom', 'head_shake', 'pounce', 'rest', 'roll', 'run', 'swim', 'walk']\n"
     ]
    }
   ],
   "source": [
    "## Integer codes for specific behaviors\n",
    "\n",
    "beh_names = sorted(set([obs_beh_to_beh_name[k] for k in obs_beh_to_beh_name]))\n",
    "beh_names.remove('unknown')\n",
    "beh_names.insert(0, 'unknown')\n",
    "\n",
    "print(beh_names)\n",
    "\n",
    "dataset_metadata['label_names'] = beh_names\n",
    "\n",
    "beh_str_to_int = {name : i for i, name in enumerate(beh_names)}\n",
    "beh_int_to_str = {i : name for i, name in enumerate(beh_names)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "f438d48f-0596-4a58-8669-358c9a4cec2d",
   "metadata": {},
   "outputs": [],
   "source": [
    "## clip_data column names\n",
    "\n",
    "clip_column_names = ['AccX', 'AccY', 'AccZ', 'Wetdry', 'individual_id', 'label']\n",
    "dataset_metadata['clip_column_names'] = clip_column_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "a560ecfd-2ff8-4aff-a58c-76bb31200b86",
   "metadata": {},
   "outputs": [],
   "source": [
    "## Save off dataset level metadata\n",
    "\n",
    "dataset_metadata_fp = os.path.join(formatted_data_dir, 'dataset_metadata.yaml')\n",
    "\n",
    "with open(dataset_metadata_fp, 'w') as file:\n",
    "    yaml.dump(dataset_metadata, file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "9181f651-b141-4e91-bfd4-a4a25f5a10c9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "31288"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# how many annotations?\n",
    "len(annotations[annotations['Bear'].isin(individuals)])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f964b042-2a2b-4690-bdd1-1685991ca77f",
   "metadata": {},
   "source": [
    "## Format Clip Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "a81e0e3f-4b1b-4957-a6f9-aca4601efe13",
   "metadata": {},
   "outputs": [],
   "source": [
    "# add annotations to data\n",
    "\n",
    "annotations['ACC_corrected_Datetime_behavior_start'] = pd.to_datetime(annotations['ACC_corrected_Datetime_behavior_start'])\n",
    "annotations['ACC_corrected_Datetime_behavior_end'] = pd.to_datetime(annotations['ACC_corrected_Datetime_behavior_end'])\n",
    "data['Datetime'] = pd.to_datetime(data['Datetime'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "7db02b94-6e8c-4469-afa6-c3739bb031a5",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Convenience function to detect nans.\n",
    "def nan_helper(y):\n",
    "    return np.isnan(y), lambda z: z.nonzero()[0]\n",
    "\n",
    "#Interpolates all nan values of given array\n",
    "def interpolate_nan(y):\n",
    "    nans, x= nan_helper(y)\n",
    "    y[nans]= np.interp(x(nans), x(~nans), y[~nans])\n",
    "    return y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "c7bd7e0b-68ed-4814-9977-284f8877a143",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "adding annotations for individual 5\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 8061/8061 [18:55<00:00,  7.10it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "processing clips for individual 5\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 39/39 [01:29<00:00,  2.29s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "adding annotations for individual 6\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4145/4145 [10:35<00:00,  6.52it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "processing clips for individual 6\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 39/39 [01:30<00:00,  2.33s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "adding annotations for individual 7\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4997/4997 [13:42<00:00,  6.07it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "processing clips for individual 7\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 39/39 [01:42<00:00,  2.62s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "adding annotations for individual 8\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 7561/7561 [15:41<00:00,  8.03it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "processing clips for individual 8\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 39/39 [01:12<00:00,  1.86s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "adding annotations for individual 9\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 6524/6524 [15:37<00:00,  6.96it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "processing clips for individual 9\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 39/39 [01:43<00:00,  2.66s/it]\n"
     ]
    }
   ],
   "source": [
    "#data['beh_label'] = 'unknown'\n",
    "data['beh_int_label'] = 0\n",
    "\n",
    "total_dur_samples = 0\n",
    "total_annotation_dur_samples = 0\n",
    "\n",
    "for individual in individuals:\n",
    "    data_sub = data[data['Bear'] == individual].copy()\n",
    "    annotations_sub = annotations[annotations['Bear'] == individual]\n",
    "    #add annotations for that individual\n",
    "    print(\"adding annotations for individual %d\" % individual)\n",
    "    for i, row in tqdm.tqdm(annotations_sub.iterrows(), total = len(annotations_sub)):\n",
    "        # Add annotations, accounting for synchronization\n",
    "        # offset = time_corrections[individual](row['ACC_corrected_Datetime_behavior_start'])\n",
    "        start = row['ACC_corrected_Datetime_behavior_start'] #+ offset\n",
    "        end = row['ACC_corrected_Datetime_behavior_end'] #+ offset\n",
    "        obs_beh = row['Behavior']\n",
    "        selected = data_sub[(data_sub['Datetime'] >= start) & (data_sub['Datetime'] <= end)].index\n",
    "        #data.loc[selected, 'beh_label'] = obs_beh_to_beh_name[obs_beh]\n",
    "        data_sub.loc[selected, 'beh_int_label'] = beh_str_to_int[obs_beh_to_beh_name[obs_beh]]\n",
    "        \n",
    "    # For each clip, create and save the data in the correct format\n",
    "    print(\"processing clips for individual %d\" % individual)\n",
    "    for clip_id in tqdm.tqdm(clip_ids):\n",
    "        individual_id = clip_id_to_individual_id[clip_id]\n",
    "        if individual_id != individual:\n",
    "            continue\n",
    "            \n",
    "        # load features\n",
    "        data_sub_clip = data_sub[data_sub['clip_id'] == clip_id].copy()\n",
    "        features = np.array(data_sub_clip[['Int_aX', 'Int_aY', 'Int_aZ']])\n",
    "        clip_dur_samples = np.shape(features)[0]\n",
    "        total_dur_samples+= clip_dur_samples\n",
    "            \n",
    "        # Water conductivity\n",
    "        wetdry = interpolate_nan(np.array(data_sub_clip['Wetdry']))\n",
    "        wetdry = np.expand_dims(wetdry, axis = -1)\n",
    "\n",
    "        per_frame_annotations = np.array(data_sub_clip['beh_int_label'])\n",
    "        per_frame_annotations = np.expand_dims(per_frame_annotations, axis = -1)\n",
    "\n",
    "        # insert individual id\n",
    "\n",
    "        per_frame_individual_id = np.full_like(per_frame_annotations, individual_id)\n",
    "\n",
    "        clip_data = np.concatenate([features, wetdry, per_frame_individual_id, per_frame_annotations], axis = 1)\n",
    "\n",
    "        clip_data_fp = os.path.join(clip_data_dir, clip_id + '.csv')\n",
    "        np.savetxt(clip_data_fp, clip_data, delimiter=\",\")\n",
    "        \n",
    "        clip_annotation_dur_samples = np.sum(per_frame_annotations > 0)\n",
    "        total_annotation_dur_samples += clip_annotation_dur_samples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "e9bba2cc-6936-4b9d-ae10-9028ba2da696",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "total duration is 3990300.312500 seconds, i.e. 1108.416753 hours\n",
      "total annotation duration is 705253.750000 seconds, i.e. 195.903819 hours\n"
     ]
    }
   ],
   "source": [
    "total_dur_sec = total_dur_samples / sr\n",
    "total_dur_hours = total_dur_sec / 3600\n",
    "print(\"total duration is %f seconds, i.e. %f hours\" % (total_dur_sec, total_dur_hours))\n",
    "\n",
    "total_annotation_dur_sec = total_annotation_dur_samples / sr\n",
    "total_annotation_dur_hours = total_annotation_dur_sec / 3600\n",
    "print(\"total annotation duration is %f seconds, i.e. %f hours\" % (total_annotation_dur_sec, total_annotation_dur_hours))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "ad0df782-6131-422c-b9bd-99710a58bf1a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0     9045281\n",
       "6     2447760\n",
       "10     432624\n",
       "2       93648\n",
       "7       36048\n",
       "4       11904\n",
       "1        8976\n",
       "9        7280\n",
       "3        3472\n",
       "8        1280\n",
       "5        1008\n",
       "Name: beh_int_label, dtype: int64"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_sub['beh_int_label'].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c9844300-e520-49a9-b975-210caa20a21f",
   "metadata": {},
   "source": [
    "## Check it all looks ok"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "717812e0-af4e-4bd0-96db-5ce83ece61c3",
   "metadata": {},
   "outputs": [],
   "source": [
    "clip_id = clip_ids[4]\n",
    "clip_data_fp = os.path.join(clip_data_dir, clip_id + '.csv')\n",
    "data = np.genfromtxt(clip_data_fp, delimiter = ',')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "77164f71-4257-43b1-ae76-49f1438c836e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1728000, 6)"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.shape(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "b1d243b7-ec2e-4bc5-8239-400cc7241953",
   "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": [
    "for i in range(np.shape(data)[1]):\n",
    "    plt.plot(data[1000:1100, i])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "ff4e9c59-9f68-4ee9-9cfd-0d1ca21b5900",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7faa5c4d2650>]"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "plt.plot(data[:, 1])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "80db1722-5a4d-4ced-ac90-58828ec4af5e",
   "metadata": {},
   "source": [
    "## Plot Dataset Summary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "c2ddf48f-b53d-465a-86cb-a155259dc0fb",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Count class representation\n",
    "\n",
    "beh_names = dataset_metadata['label_names']\n",
    "label_count = {fold : {name : 0 for name in beh_names} for fold in range(n_folds)}\n",
    "\n",
    "for clip_id in clip_ids:\n",
    "    clip_data_fp = os.path.join(clip_data_dir, clip_id + '.csv')\n",
    "    data = pd.read_csv(clip_data_fp, delimiter = ',', header = None)\n",
    "    counts = data[data.keys()[-1]].astype(int).value_counts()\n",
    "    for key in counts.keys():\n",
    "        name = beh_names[key]\n",
    "        for fold in range(n_folds):\n",
    "            if clip_id in dataset_metadata['clip_ids_per_fold'][fold]:\n",
    "                label_count[fold][name] += counts[key]\n",
    "                \n",
    "label_perc = {fold : {name : 0 for name in beh_names[1:]} for fold in range(n_folds)}\n",
    "\n",
    "for name in beh_names[1:]:\n",
    "    for fold in range(n_folds):\n",
    "        label_perc[fold][name] = float(label_count[fold][name] / sum([label_count[fold][j] for j in beh_names[1:]]))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "9898cb5a-6773-41da-b4e2-bcc6cc5f0fbd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot class representation\n",
    "\n",
    "# https://gist.github.com/thriveth/8560036#file-cbcolors-py\n",
    "CB_color_cycle = ['#377eb8', '#ff7f00', '#4daf4a',\n",
    "                  '#f781bf', '#a65628', '#984ea3',\n",
    "                  '#999999', '#e41a1c', '#dede00']\n",
    "\n",
    "keys = np.arange(len(beh_names))[1:]\n",
    "offset = 0.1\n",
    "width = 0.1\n",
    "plt.figure(figsize=(8, 6), dpi=80)\n",
    "\n",
    "for fold in range(n_folds):\n",
    "    current_offset = -(fold - n_folds // 2)*offset\n",
    "    plt.barh(keys +current_offset, [label_perc[fold][beh_names[key]] for key in keys], height=width, align='center', label = fold, color = CB_color_cycle[fold])\n",
    "    \n",
    "plt.title(\"Class representation across folds\")\n",
    "plt.yticks(keys, beh_names[1:], fontsize = 7, rotation = 60)\n",
    "plt.xlabel('Percent of known labels')\n",
    "plt.legend()\n",
    "target_fp = os.path.join(formatted_data_dir, 'class_rep.svg')\n",
    "\n",
    "plt.savefig(target_fp, format = 'svg')\n",
    "plt.show()\n",
    "\n",
    "numerical_target_fp = os.path.join(formatted_data_dir, 'class_rep.yaml')\n",
    "with open(numerical_target_fp, 'w') as file:\n",
    "    yaml.dump(label_perc, file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "88dfe3fa-445c-4806-b4b5-88175cf0c7bd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x144 with 5 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Pie chart, of known vs unknown samples \n",
    "\n",
    "labels = 'Known', 'Unknown'\n",
    "sizes = {fold : [sum([label_count[fold][j] for j in beh_names[1:]]), label_count[fold]['unknown']] for fold in range(n_folds)}\n",
    "# train_sizes = [sum([train_label_count[j] for j in beh_names[1:]]), train_label_count['unknown']]\n",
    "# val_sizes = [sum([val_label_count[j] for j in beh_names[1:]]), val_label_count['unknown']]\n",
    "# test_sizes = [sum([test_label_count[j] for j in beh_names[1:]]), test_label_count['unknown']]\n",
    "\n",
    "fig, ax = plt.subplots(1,n_folds, figsize = (3*n_folds,2))\n",
    "\n",
    "for fold in range(n_folds):\n",
    "    ax[fold].pie(sizes[fold], labels=labels, autopct='%1.1f%%', startangle=90)\n",
    "    ax[fold].axis('equal')  # Equal aspect ratio ensures that pie is drawn as a circle.\n",
    "    ax[fold].set_title(fold)\n",
    "\n",
    "target_fp = os.path.join(formatted_data_dir, 'perc_known.svg')\n",
    "\n",
    "plt.savefig(target_fp, format='svg')\n",
    "\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "165c4822-ce09-404d-b8d3-a56eebf2a1b3",
   "metadata": {},
   "source": [
    "## Temporal Statistics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "25a4416d-2ce4-4c63-8f6e-6cb9f519425d",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 39/39 [02:02<00:00,  3.15s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean overall label duration is 127.524 seconds\n",
      "Re-saving metadata to /home/jupyter/behavior_data_local/data/formatted/pagano_bears/dataset_metadata.yaml\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "####################################\n",
    "infill_max_dur_sec = 91 # If annotations are nnnnn0000000nnnnn, where 0=unknown, do we fill in 0's with n's?\n",
    "####################################\n",
    "\n",
    "# This cell runs almost as a stand alone, but requires the first few cells of this notebook to be run first\n",
    "\n",
    "import tqdm\n",
    "from pathlib import Path\n",
    "\n",
    "dataset_metadata_fp = os.path.join(formatted_data_dir, 'dataset_metadata.yaml')\n",
    "\n",
    "with open(dataset_metadata_fp, 'r') as file:\n",
    "    metadata = yaml.safe_load(file)\n",
    "\n",
    "clip_fps = list(Path(clip_data_dir).glob('*.csv'))\n",
    "\n",
    "def create_list_of_durations(x, infill_max_dur_sec, samplerate, unknown_value = 0):\n",
    "    # First pass: create a list of label durations, together with a list of the associated classes\n",
    "    current_value = 0\n",
    "    current_dur = 0\n",
    "    list_of_durs = []\n",
    "    list_of_labels = []\n",
    "    for i in x:\n",
    "        if i == current_value:\n",
    "            current_dur += 1\n",
    "        elif i != current_value:\n",
    "            list_of_durs.append(current_dur / samplerate)\n",
    "            list_of_labels.append(current_value)\n",
    "            current_dur = 1\n",
    "            current_value = i\n",
    "            \n",
    "    list_of_durs.append(current_dur)\n",
    "    list_of_labels.append(current_value)\n",
    "            \n",
    "    # repeatedly merge similar labels if they are seperated by unknowns of short duration\n",
    "    # This is mainly to deal with duty cycling of observations\n",
    "    infill_max_dur_samples = int(infill_max_dur_sec * samplerate)\n",
    "    \n",
    "    if list_of_labels[0] == unknown_value:\n",
    "        del list_of_labels[0]\n",
    "        del list_of_durs[0]\n",
    "        \n",
    "    if len(list_of_labels) > 0:\n",
    "        if list_of_labels[-1] == unknown_value:\n",
    "            del list_of_labels[-1]\n",
    "            del list_of_durs[-1]\n",
    "    \n",
    "    j = 1\n",
    "    while j<len(list_of_labels)-1:\n",
    "        if list_of_labels[j] != unknown_value:\n",
    "            j+= 1\n",
    "        elif list_of_labels[j-1] == list_of_labels[j+1] and list_of_durs[j] < infill_max_dur_samples:\n",
    "            # merge if it's a short interval of unknowns between two of the same label\n",
    "            list_of_durs[j-1] += list_of_durs[j+1]\n",
    "            del list_of_durs[j+1]\n",
    "            del list_of_durs[j]\n",
    "            del list_of_labels[j+1]\n",
    "            del list_of_labels[j]\n",
    "        else:\n",
    "            # otherwise, drop the unknown segment\n",
    "            del list_of_durs[j]\n",
    "            del list_of_labels[j]\n",
    "\n",
    "    return list_of_labels, list_of_durs # list of annotation durations, in samples.\n",
    "                             \n",
    "durs_sec_by_individual = {i : [] for i in metadata['individual_ids']}\n",
    "overall_durs_sec = []\n",
    "    \n",
    "for fp in tqdm.tqdm(clip_fps):\n",
    "    clip_id = fp.stem\n",
    "    individual_id = metadata['clip_id_to_individual_id'][clip_id]\n",
    "    clip_annotations = list(pd.read_csv(fp, header = None).values[:, -1])\n",
    "    l, d = create_list_of_durations(clip_annotations, infill_max_dur_sec, metadata['sr'])\n",
    "    durs_sec_by_individual[individual_id].extend(d)\n",
    "    overall_durs_sec.extend(d)\n",
    "    \n",
    "mean_dur_sec_by_individual = {i : float(np.mean(durs_sec_by_individual[i])) for i in durs_sec_by_individual}\n",
    "mean_overall_dur_sec = float(np.mean(overall_durs_sec))\n",
    "\n",
    "metadata['mean_dur_sec_by_individual'] = mean_dur_sec_by_individual\n",
    "metadata['mean_overall_dur_sec'] = mean_overall_dur_sec\n",
    "\n",
    "print(\"Mean overall label duration is %0.3f seconds\" % mean_overall_dur_sec)\n",
    "\n",
    "print(\"Re-saving metadata to %s\" % str(dataset_metadata_fp))\n",
    "\n",
    "with open(dataset_metadata_fp, 'w') as file:\n",
    "    yaml.dump(metadata, file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f7ca020f-a78f-4d1a-ad19-356ca2907263",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "environment": {
   "kernel": "python3",
   "name": "pytorch-gpu.1-11.m94",
   "type": "gcloud",
   "uri": "gcr.io/deeplearning-platform-release/pytorch-gpu.1-11:m94"
  },
  "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": "3.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}