Data_Science_Part_2_v1.0.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Teil 2\n",
    "\n",
    "Folgende Regressionsarten werden verwendet um eine Vorhersage des SalePrice zu erstellen:\n",
    "* Lineare Regression\n",
    "* Ridge\n",
    "* SGDRegressor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "metadata": {
    "pycharm": {
     "is_executing": false,
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.preprocessing import RobustScaler\n",
    "from sklearn.preprocessing import OneHotEncoder\n",
    "from sklearn.model_selection import cross_validate\n",
    "from sklearn.compose import make_column_transformer\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import r2_score, mean_squared_error, max_error\n",
    "from sklearn.linear_model import LinearRegression, SGDRegressor, Ridge\n",
    "\n",
    "# prevent deprecation warnings\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 169,
   "metadata": {
    "pycharm": {
     "is_executing": false,
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "# gets dataFrame from file\n",
    "def read_data(path):\n",
    "    df = pd.read_csv(path, sep=\";\")\n",
    "    return df\n",
    "\n",
    "# scaling numerical values and turning objects into numerical values\n",
    "def scale_and_convert_columns(df, model):\n",
    "    categoricalCols = []\n",
    "    for i in range(len(df.columns)):\n",
    "        if (df.dtypes[i] == object):\n",
    "            categoricalCols.append(i)\n",
    "    numericalCols = []\n",
    "    for i in range(len(df.columns)):\n",
    "        if (df.dtypes[i] == 'int64' ):\n",
    "            numericalCols.append(i)        \n",
    "            \n",
    "    preprocessor = make_column_transformer((categoricalCols, OneHotEncoder(handle_unknown='ignore')),\n",
    "                                            (numericalCols, RobustScaler()))\n",
    "    pipe = Pipeline(steps=[('preprocessor', preprocessor),\n",
    "                      ('classifier', model)])\n",
    "    return pipe\n",
    "\n",
    "# calculates R2, MSE, RMSE, MAPE and MAX* for the given regression model\n",
    "def  train_model(path, model):\n",
    "    df = read_data(path)\n",
    "    pipe = scale_and_convert_columns(df.drop('SalePrice', axis=1), model)\n",
    "    \n",
    "    x = df.drop('SalePrice', axis=1).values #dimensions\n",
    "    y = df['SalePrice'].values #prediction\n",
    "    \n",
    "    x_train, x_test, y_train, y_test = train_test_split(x, y, random_state=42)\n",
    "    pipe.fit(x_train, y_train)\n",
    "    \n",
    "    predicted = pipe.predict(x_test)\n",
    "    expected = y_test\n",
    "    mse = mean_squared_error(expected, predicted)\n",
    "    result = {'Regression Type': type(model).__name__, 'R2': r2_score(expected, predicted), 'MSE': mean_squared_error(expected, predicted), \n",
    "            'RMSE' : np.sqrt(mse),'MAPE':  np.mean(np.abs((expected-predicted)/expected) * 100), \n",
    "            'MAX*': max_error(expected, predicted)}\n",
    "    return result, model\n",
    "\n",
    "def calc_prediction(path, model):\n",
    "    df = read_data(path)\n",
    "    pipe = scale_and_convert_columns(df.drop('SalePrice', axis=1), model)\n",
    "    \n",
    "    x = df.drop('SalePrice', axis=1).values #dimensions\n",
    "    y = df['SalePrice'].values #prediction\n",
    "    \n",
    "    # neg_root_mean_squared_error somehow doesn't work\n",
    "    scores = cross_validate(pipe, x, y, cv=5, scoring=['r2', 'neg_mean_squared_error', \n",
    "                                                      'neg_median_absolute_error', 'max_error' ])\n",
    "    return {'Regression Type': type(model).__name__,'R2': scores['test_r2'].mean(), 'MSE': scores['test_neg_mean_squared_error'].mean()*-1, \n",
    "            'RMSE' : np.sqrt(scores['test_neg_mean_squared_error'].mean()*-1),'MAPE': scores['test_neg_median_absolute_error'].mean()*-1, \n",
    "            'MAX*': scores['test_max_error'].mean()*-1}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 170,
   "metadata": {
    "pycharm": {
     "is_executing": false,
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Regression Type</th>\n",
       "      <th>R2</th>\n",
       "      <th>MSE</th>\n",
       "      <th>RMSE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>MAX*</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>LinearRegression</td>\n",
       "      <td>0.894020</td>\n",
       "      <td>3.173257e+08</td>\n",
       "      <td>17813.637951</td>\n",
       "      <td>9.013587</td>\n",
       "      <td>64501.107612</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>Ridge</td>\n",
       "      <td>0.897330</td>\n",
       "      <td>3.074155e+08</td>\n",
       "      <td>17533.267550</td>\n",
       "      <td>8.789771</td>\n",
       "      <td>61505.530768</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>SGDRegressor</td>\n",
       "      <td>0.901837</td>\n",
       "      <td>2.939201e+08</td>\n",
       "      <td>17144.098171</td>\n",
       "      <td>8.384517</td>\n",
       "      <td>50957.046653</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    Regression Type        R2           MSE          RMSE      MAPE  \\\n",
       "0  LinearRegression  0.894020  3.173257e+08  17813.637951  9.013587   \n",
       "1             Ridge  0.897330  3.074155e+08  17533.267550  8.789771   \n",
       "2      SGDRegressor  0.901837  2.939201e+08  17144.098171  8.384517   \n",
       "\n",
       "           MAX*  \n",
       "0  64501.107612  \n",
       "1  61505.530768  \n",
       "2  50957.046653  "
      ]
     },
     "execution_count": 170,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x432 with 5 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#calculate the score for each model\n",
    "path = 'cleanedSet.csv'\n",
    "regressionTypes = [LinearRegression(), Ridge(), SGDRegressor()]\n",
    "results = []\n",
    "models = []\n",
    "for r in regressionTypes:\n",
    "    result, model = train_model(path, r)\n",
    "    results.append(result)\n",
    "    models.append(model)\n",
    "\n",
    "df = pd.DataFrame(data=results, columns=['Regression Type', 'R2', 'MSE', 'RMSE', 'MAPE', 'MAX*'])\n",
    "\n",
    "columns = df.columns\n",
    "fx, axes = plt.subplots(1, len(columns)-1, sharey=False, figsize=(30,6))\n",
    "for i in range(1, len(columns)):\n",
    "    # could also use barplot instead of scatterplot\n",
    "    sns.scatterplot(x=df.columns[0], y=df.columns[i], data=df, ax=axes[i-1])\n",
    "    \n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "## Ergebnis\n",
    "Generell ist bei den Ergebnissen wünschenswert, das der Bestimmtheitsgrad(R²) so hoch wie möglich ist und alle Fehlerwerte(MSE, RMSE, MAPE und MAX*) so niedrig wie möglich.\n",
    "\n",
    "Es wurden über 10 verschiedene Regressionsarten von scikit learn getestet und drei der besten (LinearRegression, Ridge und SGDRegressor) für einen genauen vergleich ausgewählt. Durch die Entfernung der Ausreiser konnte der Bestimmtheitsgrad aller 3 Regressionsarten um ca. 3-5% verbessert werden. An der Spitze aktuell der SGDRegressor mit einem Bestimmtheitsgrad von 90%. Bei genauerer Betrachtung fällt auf, das LinearRegression und Ridge sehr nahe beieinander liegen, wobei aber Ridge in allen Werten, besser abschneidet. Beide werden aber vom SGDRegressor geschlagen, vorallem im bereich maximum Error.   \n",
    "\n",
    "Anhand dieser Ergebnisse würden wir für eine Vorhersage der Verkaufspreise für den SGDRegressor entscheiden.\n",
    "\n",
    "### Analyseidee\n",
    "Um die eventuell Ergebnisse und den Prozess noch weiter zu Verbessern könnte versucht werden mehrere der Features zusammen zu fassen. Dies könnte sowohl den Lern- als auch den Vorhersageprozess beschleunigen."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "Eine Vorhersage zusätzlicher Daten kann im folgenden Codeteil vorgenommen werden. Hierfür muss nur die folgende path Variable angepasst werden."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 171,
   "metadata": {
    "pycharm": {
     "is_executing": false,
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Regression Type</th>\n",
       "      <th>R2</th>\n",
       "      <th>MSE</th>\n",
       "      <th>RMSE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>MAX*</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Empty DataFrame\n",
       "Columns: [Regression Type, R2, MSE, RMSE, MAPE, MAX*]\n",
       "Index: []"
      ]
     },
     "execution_count": 171,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "path = ''\n",
    "results = []\n",
    "if path:\n",
    "    for m in models:\n",
    "        results.append(calc_prediction(path, m))\n",
    "\n",
    "df = pd.DataFrame(data=results, columns=['Regression Type', 'R2', 'MSE', 'RMSE', 'MAPE', 'MAX*'])\n",
    "df.head()"
   ]
  }
 ],
 "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": "3.7.4"
  },
  "pycharm": {
   "stem_cell": {
    "cell_type": "raw",
    "metadata": {
     "collapsed": false
    },
    "source": []
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}