models.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Baltimore Ravens Free Agent Value Prediction Challenge\n",
    "\n",
    "Modeling tutorial\n",
    "\n",
    "## Load libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load data\n",
    "\n",
    "We use the combined dataset from the starter code."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>player_id_otc</th>\n",
       "      <th>Team</th>\n",
       "      <th>team_id</th>\n",
       "      <th>Position</th>\n",
       "      <th>Role</th>\n",
       "      <th>Years</th>\n",
       "      <th>date_signed</th>\n",
       "      <th>year_signed</th>\n",
       "      <th>start_year</th>\n",
       "      <th>Total</th>\n",
       "      <th>...</th>\n",
       "      <th>pass_snaps</th>\n",
       "      <th>receiving_snaps</th>\n",
       "      <th>passblock_snaps</th>\n",
       "      <th>run_snaps</th>\n",
       "      <th>runblock_snaps</th>\n",
       "      <th>rundefense_snaps</th>\n",
       "      <th>passrush_snaps</th>\n",
       "      <th>coverage_snaps</th>\n",
       "      <th>total_snaps</th>\n",
       "      <th>games</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>51</td>\n",
       "      <td>49ers</td>\n",
       "      <td>31</td>\n",
       "      <td>43OLB</td>\n",
       "      <td>Defense</td>\n",
       "      <td>6</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2012</td>\n",
       "      <td>2012</td>\n",
       "      <td>40400000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>387.0</td>\n",
       "      <td>555.0</td>\n",
       "      <td>180.0</td>\n",
       "      <td>1122.0</td>\n",
       "      <td>18.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>53</td>\n",
       "      <td>Vikings</td>\n",
       "      <td>24</td>\n",
       "      <td>LG</td>\n",
       "      <td>Offense</td>\n",
       "      <td>4</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2016</td>\n",
       "      <td>2016</td>\n",
       "      <td>26800000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>478.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>307.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>785.0</td>\n",
       "      <td>13.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>54</td>\n",
       "      <td>Panthers</td>\n",
       "      <td>26</td>\n",
       "      <td>P</td>\n",
       "      <td>Special</td>\n",
       "      <td>2</td>\n",
       "      <td>42792.0</td>\n",
       "      <td>2017</td>\n",
       "      <td>2017</td>\n",
       "      <td>5000000</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>74</td>\n",
       "      <td>Cowboys</td>\n",
       "      <td>18</td>\n",
       "      <td>C</td>\n",
       "      <td>Offense</td>\n",
       "      <td>2</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2016</td>\n",
       "      <td>2016</td>\n",
       "      <td>1675000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>279.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>149.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>428.0</td>\n",
       "      <td>8.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>79</td>\n",
       "      <td>Raiders</td>\n",
       "      <td>16</td>\n",
       "      <td>WR</td>\n",
       "      <td>Offense</td>\n",
       "      <td>4</td>\n",
       "      <td>NaN</td>\n",
       "      <td>2015</td>\n",
       "      <td>2016</td>\n",
       "      <td>34000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>506.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>233.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>740.0</td>\n",
       "      <td>16.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 62 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   player_id_otc      Team  team_id Position     Role Years  date_signed  \\\n",
       "0             51     49ers       31    43OLB  Defense     6          NaN   \n",
       "1             53   Vikings       24       LG  Offense     4          NaN   \n",
       "2             54  Panthers       26        P  Special     2      42792.0   \n",
       "3             74   Cowboys       18        C  Offense     2          NaN   \n",
       "4             79   Raiders       16       WR  Offense     4          NaN   \n",
       "\n",
       "   year_signed  start_year     Total  ...    pass_snaps  receiving_snaps  \\\n",
       "0         2012        2012  40400000  ...           0.0              0.0   \n",
       "1         2016        2016  26800000  ...           0.0              0.0   \n",
       "2         2017        2017   5000000  ...           NaN              NaN   \n",
       "3         2016        2016   1675000  ...           0.0              0.0   \n",
       "4         2015        2016  34000000  ...           0.0            506.0   \n",
       "\n",
       "   passblock_snaps run_snaps  runblock_snaps  rundefense_snaps  \\\n",
       "0              0.0       0.0             0.0             387.0   \n",
       "1            478.0       0.0           307.0               0.0   \n",
       "2              NaN       NaN             NaN               NaN   \n",
       "3            279.0       0.0           149.0               0.0   \n",
       "4              0.0       1.0           233.0               0.0   \n",
       "\n",
       "   passrush_snaps  coverage_snaps  total_snaps  games  \n",
       "0           555.0           180.0       1122.0   18.0  \n",
       "1             0.0             0.0        785.0   13.0  \n",
       "2             NaN             NaN          NaN    NaN  \n",
       "3             0.0             0.0        428.0    8.0  \n",
       "4             0.0             0.0        740.0   16.0  \n",
       "\n",
       "[5 rows x 62 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.read_csv('contracts_players_perf.csv', index_col=0)\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Clean for model building\n",
    "\n",
    "First, we select relevant columns that will be used for prediction. Many columns are not relevant and will be left out. Many of the columns we leave out might actually be relevant and useful, but be careful not to let any information from the contracts leak into your training procedure."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['player_id_otc', 'Team', 'team_id', 'Position', 'Role', 'Years', 'date_signed', 'year_signed', 'start_year', 'Total', 'APY', 'Guarantee', 'GPY', 'GP', 'total_guarantees', 'vested_guarantees', 'signing_bonus', 'option_bonus', 'roster_bonus', 'pergame_bonus', 'workout_bonus', 'incentives', 'FA', 'Status', 'contract_type', 'Accrued', 'height', 'weight', 'DOB', 'college', 'original_team_id', 'draft_year', 'draft_round', 'draft_pick', 'Notes', 'nfl_player_id', 'Name', 'season', 'player_id_pff', 'player', 'pos', 'overall', 'overall_rk', 'pass', 'run', 'receiving', 'passblock', 'runblock', 'passrush', 'rundefense', 'coverage', 'discipline', 'pass_snaps', 'receiving_snaps', 'passblock_snaps', 'run_snaps', 'runblock_snaps', 'rundefense_snaps', 'passrush_snaps', 'coverage_snaps', 'total_snaps', 'games']\n"
     ]
    }
   ],
   "source": [
    "print(data.columns.tolist())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "predictors = ['Position', 'Role', 'Accrued', 'height', 'weight', 'DOB', 'college', # basic info\n",
    "              'Years',  'original_team_id', 'draft_year', 'draft_round', 'draft_pick', # basic NFL history\n",
    "              'overall', 'overall_rk', # pff aggregate grades\n",
    "              'pass', 'run', 'receiving', 'passblock', 'runblock', 'passrush',\n",
    "              'rundefense', 'coverage', 'discipline', 'pass_snaps', 'receiving_snaps',\n",
    "              'passblock_snaps', 'run_snaps', 'runblock_snaps', 'rundefense_snaps', \n",
    "              'passrush_snaps', 'coverage_snaps', 'total_snaps', 'games'] # detailed stats from most recent season\n",
    "\n",
    "target = 'APY'\n",
    "\n",
    "X = data[predictors]\n",
    "Y = data[target]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A few preprocessing steps need to happen before we train a model.\n",
    "\n",
    "- Dummy variables: convert categorical variables into several binary variables\n",
    "- Imputation: meaningfully handle missing data\n",
    "\n",
    "### Dummy variables\n",
    "\n",
    "First, identify all categorical variables. These usually have type \"object\" (strings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['Position', 'Role', 'height', 'DOB', 'college', 'Years', 'overall_rk',\n",
       "       'pass', 'run', 'receiving', 'passblock', 'runblock', 'passrush',\n",
       "       'rundefense', 'coverage'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X.columns[X.dtypes=='object']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see a few of these clearly are not categorical. For example, `height` is coded as a string, which we can fix.\n",
    "\n",
    "A few others are easily converted to integers or floats without much work."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Disable copy warning - it's a weird PANDAS thing\n",
    "pd.options.mode.chained_assignment = None\n",
    "\n",
    "X = X.replace('\\\\N', np.nan)\n",
    "X['Years']      = X['Years'].replace('s', np.nan).astype('float')\n",
    "X['pass']       = X['pass'].astype('float')\n",
    "X['run']        = X['run'].astype('float')\n",
    "X['receiving']  = X['receiving'].astype('float')\n",
    "X['passblock']  = X['passblock'].astype('float')\n",
    "X['runblock']   = X['runblock'].astype('float')\n",
    "X['passrush']   = X['passrush'].astype('float')\n",
    "X['rundefense'] = X['rundefense'].astype('float')\n",
    "X['coverage']   = X['coverage'].astype('float')\n",
    "X['overall_rk'] = X['overall_rk'].astype('float')\n",
    "\n",
    "X['DOB'] = (pd.datetime(2017, 1, 1) - pd.to_datetime(X.DOB)).apply(lambda x: np.nan if pd.isnull(x) else x.days/365) # now means age\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def fixHeight(x):\n",
    "    if pd.isnull(x) or len(x)<2: return np.nan\n",
    "    x = x.split('-')\n",
    "    return int(x[0])*12 + int(x[1])\n",
    "\n",
    "X['height'] = X['height'].replace(' ', np.nan).apply(fixHeight).astype('float')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we actually make the dummy variables. Let's look at the remaining categorical variables:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Position : 33\n",
      "Role : 3\n",
      "college : 449\n"
     ]
    }
   ],
   "source": [
    "for col in X.columns[X.dtypes=='object']:\n",
    "    print(col, ':', len(X[col].value_counts()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For `college`, the number of features would be unmanageable if we make one for each of the 449 schools. We will just use the top 25 colleges instead."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "X.college[X.college.isin(X.college.value_counts()[25:].index)] = np.nan"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Finally, make the dummy variables.\n",
    "X = pd.concat([X, pd.get_dummies(X[['Position', 'Role', 'college']])], axis=1)\n",
    "\n",
    "X = X.drop(['Position', 'Role', 'college'], axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Imputation\n",
    "\n",
    "One common approach is to replace missing values with the mean value for that feature. However, sometimes it is useful to \"remember\" which missing values were imputed, so we also add a second column to show which values were imputed."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "for col in X.columns:\n",
    "    if X[col].isnull().any():\n",
    "        X[col+'_nan']= X[col].isnull()\n",
    "        X[col] = X[col].fillna(X[col].mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Separate Train/Validation/Test Sets\n",
    "\n",
    "We are going to train our model on a fraction of the dataset and validate it on the rest. Additionally, for the first round of the competition, we will submit predictions for the 2017 season to the Kaggle server. Our train/val/test split will look like this:\n",
    "\n",
    "- Train: 2006-2015\n",
    "- Val: 2016\n",
    "- Test: 2017\n",
    "\n",
    "We don't have many records of contracts from earlier than 2016, so you will want to re-train your models before making your final submissions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "# training examples: 552\n",
      "# val examples: 698\n",
      "# test examples: 1615\n"
     ]
    }
   ],
   "source": [
    "Xtrain, Ytrain = X[data.year_signed < 2016],  Y[data.year_signed < 2016]/1000000\n",
    "Xval,   Yval   = X[data.year_signed == 2016], Y[data.year_signed == 2016]/1000000\n",
    "Xtest,  _      = X[data.year_signed == 2017], Y[data.year_signed == 2017]/1000000 # throw away APYs\n",
    "\n",
    "print('# training examples:', len(Xtrain))\n",
    "print('# val examples:', len(Xval))\n",
    "print('# test examples:', len(Xtest))\n",
    "\n",
    "# Save train, val, and test sets\n",
    "train = Xtrain.copy(); train['APY_millions'] = Ytrain\n",
    "val   = Xval.copy();   val['APY_millions']   = Yval\n",
    "test  = Xtest.copy();\n",
    "\n",
    "train.to_csv('train.csv')\n",
    "val.to_csv('val.csv')\n",
    "test.to_csv('test.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Build Random Forest model\n",
    "\n",
    "Using our training set, we will train an ensemble of several thousand decision trees. These trees will each independently predict the APY for each player. We then average these predictions together, to get a prediction for the ensemble.\n",
    "\n",
    "For more detailed background on decision trees and forests, check out [MDST's other tutorials](https://github.com/MichiganDataScienceTeam/tutorials/blob/master/tree-based-models/trees.ipynb)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.ensemble import RandomForestRegressor\n",
    "from sklearn.metrics import mean_squared_error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "RandomForestRegressor(bootstrap=True, criterion='mse', max_depth=None,\n",
       "           max_features='auto', max_leaf_nodes=None,\n",
       "           min_impurity_split=1e-07, min_samples_leaf=1,\n",
       "           min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
       "           n_estimators=5000, n_jobs=-1, oob_score=False,\n",
       "           random_state=None, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Initialize the forest. We just set the number of trees to 5000 and leave the rest of the options the same.\n",
    "rf = RandomForestRegressor(n_estimators=5000, n_jobs=-1)\n",
    "\n",
    "# Fit the model to training data\n",
    "rf.fit(Xtrain, Ytrain)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Make predictions and evaluate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Make predictions on train, val, test\n",
    "Ytrain_hat = rf.predict(Xtrain)\n",
    "Yval_hat = rf.predict(Xval)\n",
    "Ytest_hat = rf.predict(Xtest)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Evaluate our model on train and validation\n",
    "mse_train = mean_squared_error(Ytrain_hat, Ytrain)\n",
    "mse_val = mean_squared_error(Yval_hat, Yval)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "------ Train ------\n",
      "RF:    MSE = 00.423\n",
      "---- Validation ---\n",
      "Zeros: MSE = 27.301\n",
      "Mean:  MSE = 17.363\n",
      "RF:    MSE = 03.008\n"
     ]
    }
   ],
   "source": [
    "# Compare with some simple baselines\n",
    "print('------ Train ------')\n",
    "print('RF:    MSE = %06.3f'%mse_train)\n",
    "print('---- Validation ---')\n",
    "print('Zeros: MSE = %06.3f'%mean_squared_error([0]*len(Ytrain), Ytrain))\n",
    "print('Mean:  MSE = %06.3f'%mean_squared_error([Ytrain.mean()]*len(Ytrain), Ytrain))\n",
    "print('RF:    MSE = %06.3f'%mse_val)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So, our model seems to be doing OK. We achieved an MSE of 3.031, which is better than predicting all zeros or just the average APY. However, our training set performance is much better, than our validation set performance, so we might be overfitting."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x116ff89b0>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEKCAYAAAAfGVI8AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYRgZzv8Di1c/b3ChkNUAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x116edc080>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Visualize the predictions\n",
    "\n",
    "plt.scatter(Yval_hat, Yval)\n",
    "plt.plot([0, 25], [0, 25], 'r-')\n",
    "plt.xlabel('Predicted APY ($10Million)')\n",
    "plt.ylabel('Actual APY ($10Million)')\n",
    "plt.legend([ 'perfect', 'RF'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Export predictions for Kaggle\n",
    "\n",
    "Check out the [Kaggle page](https://www.kaggle.com/c/nfl-free-agent-value) to submit to the first round of the competition!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "pd.DataFrame({'Contract_ID': Xtest.index, 'APY_millions':Ytest_hat}).to_csv('rf_submission.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}