ensemble-folds-with-median-0-153.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "25bc44b9",
   "metadata": {
    "papermill": {
     "duration": 0.01119,
     "end_time": "2021-10-03T10:08:44.877160",
     "exception": false,
     "start_time": "2021-10-03T10:08:44.865970",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Ensemble Folds with MEDIAN - [0.153]\n",
    "This notebook is a fork of [Zhangxin's][1] notebook [here][2]. The purpose of this notebook is to demonstrate using **median** (instead of mean) to ensemble fold model predictions. Additionally it demonstrates \"rounding\" predictions to match the discrete target distribution observed in train data.\n",
    "\n",
    "Since the [competition metric][3] is MAE, using **median** is best. If the competition metric were RSME, then **mean** would be best.\n",
    "\n",
    "The original notebook ensembles its 10 fold models with **mean** (instead of median) and achieves **LB 0.157**. By using **median** we boost the LB by **+0.002** and by using \"rounding\", we boost LB by **+0.002** too. Both of these tricks together boost LB a total **+0.004**!\n",
    "\n",
    "[1]: https://www.kaggle.com/tenffe\n",
    "[2]: https://www.kaggle.com/tenffe/finetune-of-tensorflow-bidirectional-lstm\n",
    "[3]: https://www.kaggle.com/c/ventilator-pressure-prediction/overview/evaluation"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "df805128",
   "metadata": {
    "papermill": {
     "duration": 0.009712,
     "end_time": "2021-10-03T10:08:44.897653",
     "exception": false,
     "start_time": "2021-10-03T10:08:44.887941",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Train Targets are Discrete - EDA\n",
    "The discovery that train targets are discrete was first published in discussion [here][1], then [here][2]\n",
    "\n",
    "[1]: https://www.kaggle.com/c/ventilator-pressure-prediction/discussion/275897\n",
    "[2]: https://www.kaggle.com/c/ventilator-pressure-prediction/discussion/276083"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "dd50c0d7",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-03T10:08:44.999893Z",
     "iopub.status.busy": "2021-10-03T10:08:44.999157Z",
     "iopub.status.idle": "2021-10-03T10:08:56.565031Z",
     "shell.execute_reply": "2021-10-03T10:08:56.564213Z",
     "shell.execute_reply.started": "2021-10-03T09:33:23.884894Z"
    },
    "papermill": {
     "duration": 11.657592,
     "end_time": "2021-10-03T10:08:56.565216",
     "exception": false,
     "start_time": "2021-10-03T10:08:44.907624",
     "status": "completed"
    },
    "tags": []
   },
   "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"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Max pressure = 64.82099173863948 Min pressure = -1.8957442945646408\n"
     ]
    }
   ],
   "source": [
    "import cupy, cudf, matplotlib.pyplot as plt\n",
    "train_gf = cudf.read_csv('../input/ventilator-pressure-prediction/train.csv')\n",
    "\n",
    "plt.title('Histogram of Train Pressures',size=14)\n",
    "plt.hist(train_gf.sample(100_000).pressure.to_array(),bins=100)\n",
    "plt.show()\n",
    "print('Max pressure =',train_gf.pressure.max(), 'Min pressure =',train_gf.pressure.min())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "f4c2a79a",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-03T10:08:56.593039Z",
     "iopub.status.busy": "2021-10-03T10:08:56.592326Z",
     "iopub.status.idle": "2021-10-03T10:08:57.180611Z",
     "shell.execute_reply": "2021-10-03T10:08:57.179984Z",
     "shell.execute_reply.started": "2021-10-03T09:49:40.737695Z"
    },
    "papermill": {
     "duration": 0.603686,
     "end_time": "2021-10-03T10:08:57.180767",
     "exception": false,
     "start_time": "2021-10-03T10:08:56.577081",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The first 25 unique pressures...\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([-1.89574429, -1.82544215, -1.75514   , -1.68483786, -1.61453571,\n",
       "       -1.54423357, -1.47393142, -1.40362928, -1.33332713, -1.26302499,\n",
       "       -1.19272284, -1.12242069, -1.05211855, -0.9818164 , -0.91151426,\n",
       "       -0.84121211, -0.77090997, -0.70060782, -0.63030568, -0.56000353,\n",
       "       -0.48970139, -0.41939924, -0.34909709, -0.27879495, -0.2084928 ])"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_pressure = cupy.sort( train_gf.pressure.unique().values )\n",
    "print('The first 25 unique pressures...')\n",
    "PRESSURE_MIN = all_pressure[0].item()\n",
    "PRESSURE_MAX = all_pressure[-1].item()\n",
    "all_pressure[:25]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "d164a3ed",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-03T10:08:57.209133Z",
     "iopub.status.busy": "2021-10-03T10:08:57.206539Z",
     "iopub.status.idle": "2021-10-03T10:08:57.673636Z",
     "shell.execute_reply": "2021-10-03T10:08:57.674089Z",
     "shell.execute_reply.started": "2021-10-03T09:49:43.050213Z"
    },
    "papermill": {
     "duration": 0.481279,
     "end_time": "2021-10-03T10:08:57.674245",
     "exception": false,
     "start_time": "2021-10-03T10:08:57.192966",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The differences between first 25 pressures...\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215,\n",
       "       0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215,\n",
       "       0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215,\n",
       "       0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215,\n",
       "       0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215])"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print('The differences between first 25 pressures...')\n",
    "PRESSURE_STEP = ( all_pressure[1] - all_pressure[0] ).item()\n",
    "all_pressure[1:26] - all_pressure[:25]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "71f13a1d",
   "metadata": {
    "papermill": {
     "duration": 0.01257,
     "end_time": "2021-10-03T10:08:57.699435",
     "exception": false,
     "start_time": "2021-10-03T10:08:57.686865",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Load Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "2c7e3fb3",
   "metadata": {
    "_cell_guid": "3e5a0bd1-3e22-4b2c-a565-68985e55f95e",
    "_kg_hide-input": true,
    "_uuid": "e331dbcc-0346-4019-9ff6-b890154a878b",
    "execution": {
     "iopub.execute_input": "2021-10-03T10:08:57.730576Z",
     "iopub.status.busy": "2021-10-03T10:08:57.729891Z",
     "iopub.status.idle": "2021-10-03T10:09:12.165369Z",
     "shell.execute_reply": "2021-10-03T10:09:12.164460Z",
     "shell.execute_reply.started": "2021-10-03T09:33:35.358530Z"
    },
    "jupyter": {
     "outputs_hidden": false
    },
    "papermill": {
     "duration": 14.454263,
     "end_time": "2021-10-03T10:09:12.165542",
     "exception": false,
     "start_time": "2021-10-03T10:08:57.711279",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import optuna\n",
    "\n",
    "import os \n",
    "os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' \n",
    "# https://www.kaggle.com/c/ventilator-pressure-prediction/discussion/274717\n",
    "\n",
    "import tensorflow as tf, gc\n",
    "from tensorflow import keras\n",
    "import tensorflow.keras.backend as K\n",
    "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint\n",
    "from tensorflow.keras.callbacks import LearningRateScheduler, ReduceLROnPlateau\n",
    "from tensorflow.keras.optimizers.schedules import ExponentialDecay\n",
    "\n",
    "from sklearn.metrics import mean_absolute_error as mae\n",
    "from sklearn.preprocessing import RobustScaler, normalize\n",
    "from sklearn.model_selection import train_test_split, GroupKFold, KFold\n",
    "\n",
    "from IPython.display import display\n",
    "\n",
    "DEBUG = False\n",
    "TRAIN_MODEL = False\n",
    "\n",
    "train = pd.read_csv('../input/ventilator-pressure-prediction/train.csv')\n",
    "test = pd.read_csv('../input/ventilator-pressure-prediction/test.csv')\n",
    "submission = pd.read_csv('../input/ventilator-pressure-prediction/sample_submission.csv')\n",
    "\n",
    "if DEBUG:\n",
    "    train = train[:80*1000]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "60b94431",
   "metadata": {
    "papermill": {
     "duration": 0.01295,
     "end_time": "2021-10-03T10:09:12.190776",
     "exception": false,
     "start_time": "2021-10-03T10:09:12.177826",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Engineer Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "d15eea2f",
   "metadata": {
    "_cell_guid": "13a36b46-7067-4b29-aad3-7e3b15e8415b",
    "_kg_hide-input": true,
    "_uuid": "dc41dbf2-f199-4b9d-bbd9-bf6084162b47",
    "execution": {
     "iopub.execute_input": "2021-10-03T10:09:12.240584Z",
     "iopub.status.busy": "2021-10-03T10:09:12.239799Z",
     "iopub.status.idle": "2021-10-03T10:10:11.616095Z",
     "shell.execute_reply": "2021-10-03T10:10:11.615600Z",
     "shell.execute_reply.started": "2021-10-03T09:33:49.351947Z"
    },
    "jupyter": {
     "outputs_hidden": false
    },
    "papermill": {
     "duration": 59.413634,
     "end_time": "2021-10-03T10:10:11.616236",
     "exception": false,
     "start_time": "2021-10-03T10:09:12.202602",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "def add_features(df):\n",
    "    df['area'] = df['time_step'] * df['u_in']\n",
    "    df['area'] = df.groupby('breath_id')['area'].cumsum()\n",
    "    \n",
    "    df['u_in_cumsum'] = (df['u_in']).groupby(df['breath_id']).cumsum()\n",
    "    \n",
    "    df['u_in_lag1'] = df.groupby('breath_id')['u_in'].shift(1)\n",
    "    df['u_out_lag1'] = df.groupby('breath_id')['u_out'].shift(1)\n",
    "    df['u_in_lag_back1'] = df.groupby('breath_id')['u_in'].shift(-1)\n",
    "    df['u_out_lag_back1'] = df.groupby('breath_id')['u_out'].shift(-1)\n",
    "    df['u_in_lag2'] = df.groupby('breath_id')['u_in'].shift(2)\n",
    "    df['u_out_lag2'] = df.groupby('breath_id')['u_out'].shift(2)\n",
    "    df['u_in_lag_back2'] = df.groupby('breath_id')['u_in'].shift(-2)\n",
    "    df['u_out_lag_back2'] = df.groupby('breath_id')['u_out'].shift(-2)\n",
    "    df['u_in_lag3'] = df.groupby('breath_id')['u_in'].shift(3)\n",
    "    df['u_out_lag3'] = df.groupby('breath_id')['u_out'].shift(3)\n",
    "    df['u_in_lag_back3'] = df.groupby('breath_id')['u_in'].shift(-3)\n",
    "    df['u_out_lag_back3'] = df.groupby('breath_id')['u_out'].shift(-3)\n",
    "    df['u_in_lag4'] = df.groupby('breath_id')['u_in'].shift(4)\n",
    "    df['u_out_lag4'] = df.groupby('breath_id')['u_out'].shift(4)\n",
    "    df['u_in_lag_back4'] = df.groupby('breath_id')['u_in'].shift(-4)\n",
    "    df['u_out_lag_back4'] = df.groupby('breath_id')['u_out'].shift(-4)\n",
    "    df = df.fillna(0)\n",
    "    \n",
    "    df['breath_id__u_in__max'] = df.groupby(['breath_id'])['u_in'].transform('max')\n",
    "    df['breath_id__u_out__max'] = df.groupby(['breath_id'])['u_out'].transform('max')\n",
    "    \n",
    "    df['u_in_diff1'] = df['u_in'] - df['u_in_lag1']\n",
    "    df['u_out_diff1'] = df['u_out'] - df['u_out_lag1']\n",
    "    df['u_in_diff2'] = df['u_in'] - df['u_in_lag2']\n",
    "    df['u_out_diff2'] = df['u_out'] - df['u_out_lag2']\n",
    "    \n",
    "    df['breath_id__u_in__diffmax'] = df.groupby(['breath_id'])['u_in'].transform('max') - df['u_in']\n",
    "    df['breath_id__u_in__diffmean'] = df.groupby(['breath_id'])['u_in'].transform('mean') - df['u_in']\n",
    "    \n",
    "    df['breath_id__u_in__diffmax'] = df.groupby(['breath_id'])['u_in'].transform('max') - df['u_in']\n",
    "    df['breath_id__u_in__diffmean'] = df.groupby(['breath_id'])['u_in'].transform('mean') - df['u_in']\n",
    "    \n",
    "    df['u_in_diff3'] = df['u_in'] - df['u_in_lag3']\n",
    "    df['u_out_diff3'] = df['u_out'] - df['u_out_lag3']\n",
    "    df['u_in_diff4'] = df['u_in'] - df['u_in_lag4']\n",
    "    df['u_out_diff4'] = df['u_out'] - df['u_out_lag4']\n",
    "    df['cross']= df['u_in']*df['u_out']\n",
    "    df['cross2']= df['time_step']*df['u_out']\n",
    "    \n",
    "    df['R'] = df['R'].astype(str)\n",
    "    df['C'] = df['C'].astype(str)\n",
    "    df['R__C'] = df[\"R\"].astype(str) + '__' + df[\"C\"].astype(str)\n",
    "    df = pd.get_dummies(df)\n",
    "    return df\n",
    "\n",
    "train = add_features(train)\n",
    "test = add_features(test)\n",
    "\n",
    "targets = train[['pressure']].to_numpy().reshape(-1, 80)\n",
    "train.drop(['pressure', 'id', 'breath_id'], axis=1, inplace=True)\n",
    "test = test.drop(['id', 'breath_id'], axis=1)\n",
    "\n",
    "RS = RobustScaler()\n",
    "train = RS.fit_transform(train)\n",
    "test = RS.transform(test)\n",
    "\n",
    "train = train.reshape(-1, 80, train.shape[-1])\n",
    "test = test.reshape(-1, 80, train.shape[-1])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a7348a42",
   "metadata": {
    "papermill": {
     "duration": 0.012176,
     "end_time": "2021-10-03T10:10:11.640894",
     "exception": false,
     "start_time": "2021-10-03T10:10:11.628718",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Infer Test Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "06a764c2",
   "metadata": {
    "_kg_hide-input": true,
    "execution": {
     "iopub.execute_input": "2021-10-03T10:10:11.677418Z",
     "iopub.status.busy": "2021-10-03T10:10:11.676881Z",
     "iopub.status.idle": "2021-10-03T10:17:28.823693Z",
     "shell.execute_reply": "2021-10-03T10:17:28.823177Z",
     "shell.execute_reply.started": "2021-10-03T09:35:33.913574Z"
    },
    "papermill": {
     "duration": 437.170882,
     "end_time": "2021-10-03T10:17:28.823848",
     "exception": false,
     "start_time": "2021-10-03T10:10:11.652966",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--------------- > Fold 1 < ---------------\n",
      "50/50 - 38s\n",
      "--------------- > Fold 2 < ---------------\n",
      "50/50 - 32s\n",
      "--------------- > Fold 3 < ---------------\n",
      "50/50 - 32s\n",
      "--------------- > Fold 4 < ---------------\n",
      "50/50 - 32s\n",
      "--------------- > Fold 5 < ---------------\n",
      "50/50 - 32s\n",
      "--------------- > Fold 6 < ---------------\n",
      "50/50 - 32s\n",
      "--------------- > Fold 7 < ---------------\n",
      "50/50 - 32s\n",
      "--------------- > Fold 8 < ---------------\n",
      "50/50 - 32s\n",
      "--------------- > Fold 9 < ---------------\n",
      "50/50 - 32s\n",
      "--------------- > Fold 10 < ---------------\n",
      "50/50 - 32s\n"
     ]
    }
   ],
   "source": [
    "EPOCH = 300\n",
    "BATCH_SIZE = 1024\n",
    "NUM_FOLDS = 10\n",
    "\n",
    "gpu_strategy = tf.distribute.get_strategy()\n",
    "\n",
    "with gpu_strategy.scope():\n",
    "    kf = KFold(n_splits=NUM_FOLDS, shuffle=True, random_state=2021)\n",
    "    test_preds = []\n",
    "    for fold, (train_idx, test_idx) in enumerate(kf.split(train, targets)):\n",
    "        K.clear_session()\n",
    "        \n",
    "        print('-'*15, '>', f'Fold {fold+1}', '<', '-'*15)\n",
    "        X_train, X_valid = train[train_idx], train[test_idx]\n",
    "        y_train, y_valid = targets[train_idx], targets[test_idx]\n",
    "        \n",
    "        checkpoint_filepath = f\"folds{fold}.hdf5\"\n",
    "        if TRAIN_MODEL:\n",
    "            model = keras.models.Sequential([\n",
    "                keras.layers.Input(shape=train.shape[-2:]),\n",
    "                keras.layers.Bidirectional(keras.layers.LSTM(1024, return_sequences=True)),\n",
    "                keras.layers.Bidirectional(keras.layers.LSTM(512, return_sequences=True)),\n",
    "                keras.layers.Bidirectional(keras.layers.LSTM(256, return_sequences=True)),\n",
    "                keras.layers.Bidirectional(keras.layers.LSTM(128, return_sequences=True)),\n",
    "                keras.layers.Dense(128, activation='selu'),\n",
    "                keras.layers.Dense(1),\n",
    "            ])\n",
    "            model.compile(optimizer=\"adam\", loss=\"mae\")\n",
    "\n",
    "            lr = ReduceLROnPlateau(monitor=\"val_loss\", factor=0.5, patience=10, verbose=1)\n",
    "            es = EarlyStopping(monitor=\"val_loss\", patience=60, verbose=1, mode=\"min\", restore_best_weights=True)\n",
    "            sv = keras.callbacks.ModelCheckpoint(\n",
    "                checkpoint_filepath, monitor='val_loss', verbose=1, save_best_only=True,\n",
    "                save_weights_only=False, mode='auto', save_freq='epoch',\n",
    "                options=None\n",
    "            )\n",
    "            model.fit(X_train, y_train, validation_data=(X_valid, y_valid), epochs=EPOCH, batch_size=BATCH_SIZE, callbacks=[lr, es, sv])\n",
    "        else:\n",
    "            model = keras.models.load_model('../input/finetune-of-tensorflow-bidirectional-lstm/'+checkpoint_filepath)\n",
    "            \n",
    "        test_preds.append(model.predict(test, batch_size=BATCH_SIZE, verbose=2).squeeze().reshape(-1, 1).squeeze())\n",
    "        \n",
    "        del model, X_train, X_valid; gc.collect()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "46fc8bcd",
   "metadata": {
    "papermill": {
     "duration": 0.016761,
     "end_time": "2021-10-03T10:17:28.858021",
     "exception": false,
     "start_time": "2021-10-03T10:17:28.841260",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Post Process with Median and Round"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e643a0cf",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-03T10:17:28.899009Z",
     "iopub.status.busy": "2021-10-03T10:17:28.897772Z",
     "iopub.status.idle": "2021-10-03T10:18:02.659490Z",
     "shell.execute_reply": "2021-10-03T10:18:02.658522Z",
     "shell.execute_reply.started": "2021-10-03T09:50:00.850057Z"
    },
    "papermill": {
     "duration": 33.784586,
     "end_time": "2021-10-03T10:18:02.659632",
     "exception": false,
     "start_time": "2021-10-03T10:17:28.875046",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# ENSEMBLE FOLDS WITH MEAN\n",
    "submission[\"pressure\"] = sum(test_preds)/NUM_FOLDS\n",
    "submission.to_csv('submission_mean_LB157.csv', index=False)\n",
    "\n",
    "# ENSEMBLE FOLDS WITH MEDIAN\n",
    "submission[\"pressure\"] = np.median(np.vstack(test_preds),axis=0)\n",
    "submission.to_csv('submission_median_LB155.csv', index=False)\n",
    "\n",
    "# ENSEMBLE FOLDS WITH MEDIAN AND ROUND PREDICTIONS\n",
    "submission[\"pressure\"] =\\\n",
    "    np.round( (submission.pressure - PRESSURE_MIN)/PRESSURE_STEP ) * PRESSURE_STEP + PRESSURE_MIN\n",
    "submission.pressure = np.clip(submission.pressure, PRESSURE_MIN, PRESSURE_MAX)\n",
    "submission.to_csv('submission_median_round_LB153.csv', index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "63fa8c43",
   "metadata": {
    "papermill": {
     "duration": 0.017162,
     "end_time": "2021-10-03T10:18:02.694771",
     "exception": false,
     "start_time": "2021-10-03T10:18:02.677609",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# After PostProcess, Test Targets are Discrete - EDA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "72af4299",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-03T10:18:02.759808Z",
     "iopub.status.busy": "2021-10-03T10:18:02.758452Z",
     "iopub.status.idle": "2021-10-03T10:18:03.200444Z",
     "shell.execute_reply": "2021-10-03T10:18:03.200009Z",
     "shell.execute_reply.started": "2021-10-03T09:51:08.530740Z"
    },
    "papermill": {
     "duration": 0.488345,
     "end_time": "2021-10-03T10:18:03.200567",
     "exception": false,
     "start_time": "2021-10-03T10:18:02.712222",
     "status": "completed"
    },
    "tags": []
   },
   "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"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Max pressure = 64.82099173863328 Min pressure = -1.7551400036622216\n"
     ]
    }
   ],
   "source": [
    "plt.title('Histogram of Test Pressures',size=14)\n",
    "plt.hist(submission.sample(10_000).pressure.values, bins=100)\n",
    "plt.show()\n",
    "print('Max pressure =',submission.pressure.max(), 'Min pressure =',submission.pressure.min())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "2fef575d",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-10-03T10:18:03.243720Z",
     "iopub.status.busy": "2021-10-03T10:18:03.242896Z",
     "iopub.status.idle": "2021-10-03T10:18:03.287092Z",
     "shell.execute_reply": "2021-10-03T10:18:03.287551Z",
     "shell.execute_reply.started": "2021-10-03T10:03:02.308093Z"
    },
    "papermill": {
     "duration": 0.067768,
     "end_time": "2021-10-03T10:18:03.287690",
     "exception": false,
     "start_time": "2021-10-03T10:18:03.219922",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The differences between first 25 test pressures...\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215,\n",
       "       0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215,\n",
       "       0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215,\n",
       "       0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215,\n",
       "       0.07030215, 0.07030215, 0.07030215, 0.07030215, 0.07030215])"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_pressure = np.sort( submission.pressure.unique() )\n",
    "print('The differences between first 25 test pressures...')\n",
    "all_pressure[1:26] - all_pressure[:25]"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  },
  "papermill": {
   "default_parameters": {},
   "duration": 567.950998,
   "end_time": "2021-10-03T10:18:06.072529",
   "environment_variables": {},
   "exception": null,
   "input_path": "__notebook__.ipynb",
   "output_path": "__notebook__.ipynb",
   "parameters": {},
   "start_time": "2021-10-03T10:08:38.121531",
   "version": "2.3.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}