LoanDecision.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from time import time\n",
    "import pandas as pd\n",
    "\n",
    "from sklearn import preprocessing\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "\n",
    "def clean_data(input_df):\n",
    "    \"\"\"Cleans dataframe data related to loan. Typically \n",
    "    replaces the NA values with default values and encodes\n",
    "    string values to numeric values For Ex: M,F gets \n",
    "    transformed to 0,1\n",
    "\n",
    "    Keyword arguments:\n",
    "    input_df -- The loan dataframe created out of file\n",
    " \n",
    "    \"\"\"\n",
    "    \n",
    "    #Fill missing values with default values. These default values are chosen so that\n",
    "    #bank would reject the loan with these default values\n",
    "    input_df['Gender'].fillna('M', inplace=True)\n",
    "    input_df['Married'].fillna('No', inplace=True)\n",
    "    input_df['Dependents'].fillna('0', inplace=True)\n",
    "    input_df['Education'].fillna('Not Graduate', inplace=True)\n",
    "    input_df['Self_Employed'].fillna('Yes', inplace=True)\n",
    "    input_df['ApplicantIncome'].fillna('0', inplace=True)\n",
    "    input_df['CoapplicantIncome'].fillna('0', inplace=True)\n",
    "    input_df['LoanAmount'].fillna('0', inplace=True)\n",
    "    input_df['Loan_Amount_Term'].fillna('360', inplace=True)\n",
    "    input_df['Credit_History'].fillna('0', inplace=True)\n",
    "    input_df['Property_Area'].fillna('Urban', inplace=True)\n",
    "  \n",
    "    le = preprocessing.LabelEncoder()\n",
    "\n",
    "    #Transform character values to numeric values\n",
    "    input_df['Gender'] = le.fit_transform(input_df['Gender'])\n",
    "    input_df['Married'] = le.fit_transform(input_df['Married'])\n",
    "    input_df['Education'] = le.fit_transform(input_df['Education'])\n",
    "    input_df['Self_Employed'] = le.fit_transform(input_df['Self_Employed'])\n",
    "    input_df['Property_Area'] = le.fit_transform(input_df['Property_Area'])\n",
    "      \n",
    "    #Clean dependents with values like 3+ etc\n",
    "    input_df['Dependents'] = [ int(val.replace('+','')) for val in input_df['Dependents']]  \n",
    "    \n",
    "\n",
    "def pre_process_training_data():\n",
    "    \"\"\"Pre processes the training data for loan data set \n",
    "    Merges the training features dataset with labels\n",
    "    We will not use absolute values of the loan amount and incomes \n",
    "    for training. Rather loan amount % is calculated w.r.t \n",
    "    total income of the customer. This will be fed to the algorithm. \n",
    "\n",
    "    Keyword arguments:\n",
    "    None\n",
    " \n",
    "    \"\"\"\n",
    " \n",
    "    df_loandata = pd.read_csv(\"train_data.csv\")\n",
    "      \n",
    "    clean_data(df_loandata)\n",
    "  \n",
    "    #Calculate the % of loan amount w.r.t total income\n",
    "    df_loandata['LoanAmountPct'] = [ (float(loanAmt)*100/(float(appIncom) + float(coAppIncome)))  for appIncom, coAppIncome, loanAmt in zip(df_loandata['ApplicantIncome'], df_loandata['CoapplicantIncome'], df_loandata['LoanAmount'] ) ]  \n",
    "     \n",
    "    #Assign correct value to loan status based on Y or N\n",
    "    df_loandata['Loan_Status'] = [ 1 if val == 'Y' else 0 for val in df_loandata['Loan_Status']]  \n",
    "    \n",
    "    #Drop amounts that are no longer relevant (loan_id, loan related amounts)\n",
    "    df_loandata.drop(df_loandata.columns[[0, 6, 7, 8]], axis=1, inplace=True) \n",
    "    \n",
    "    #Re arrange columns so that the loan status will come at the end\n",
    "    colnames = df_loandata.columns.tolist()\n",
    "    colnames = colnames[:-2] + colnames[-1:] + colnames[-2:-1]\n",
    "    df_loandata = df_loandata[colnames]\n",
    "      \n",
    "    #Get X (Input) and Y(Output) values  \n",
    "    X = df_loandata.iloc[:,0:-1]\n",
    "    Y = df_loandata.iloc[:, -1]\n",
    "    \n",
    "    #Preparte training and test data\n",
    "    features_train, features_test, labels_train, labels_test = train_test_split(X, Y, test_size=0.1, random_state=42)\n",
    "  \n",
    "    return features_train, features_test, labels_train, labels_test\n",
    " \n",
    "def train_model(features_train, labels_train):\n",
    "    \"\"\"Trains the model for loans\n",
    "    We will use random forest that fares better than\n",
    "    decision tree algorithm\n",
    "\n",
    "    Keyword arguments:\n",
    "    features_train -- The input features of the training data set\n",
    "    labels_train -- The label for each of the training observations\n",
    "    \"\"\"\n",
    " \n",
    "    clf = RandomForestClassifier(n_estimators=9,  max_features=None, min_samples_split=7 ,  random_state=0 )\n",
    "\n",
    "    clf = clf.fit(features_train, labels_train)\n",
    "     \n",
    "    score = clf.score(features_test, labels_test)\n",
    "    \n",
    "    print(\"Training Score %.2f %% \" % (score*100))    \n",
    "\n",
    "    return clf\n",
    "\n",
    "def pre_process_test_data():\n",
    "\n",
    "    \"\"\"Pre process the test data\n",
    "    We will not use absolute values of the loan amount and incomes \n",
    "    for training. Rather loan amount % is calculated w.r.t \n",
    "    total income of the customer. This will be fed to the algorithm. \n",
    "     \n",
    "    Keyword arguments:\n",
    "    None\n",
    "    \"\"\"\n",
    "    df_loan_test_data = pd.read_csv(\"test_data.csv\")\n",
    "    \n",
    "    clean_data(df_loan_test_data)\n",
    "  \n",
    "    #Calculate the % of loan amount w.r.t total income\n",
    "    df_loan_test_data['LoanAmountPct'] = [ (float(loanAmt)*100/(float(appIncom) + float(coAppIncome))) \n",
    "                                            for appIncom, coAppIncome, loanAmt in\n",
    "                                              zip(df_loan_test_data['ApplicantIncome'], df_loan_test_data['CoapplicantIncome'], df_loan_test_data['LoanAmount']  ) ]  \n",
    "    \n",
    "    loan_numbers = df_loan_test_data.ix[:,0]\n",
    "    #Drop amounts that are no longer relevant\n",
    "    df_loan_test_data.drop(df_loan_test_data.columns[[0, 6, 7, 8]], axis=1, inplace=True) \n",
    "  \n",
    "    #Get  Input  values  \n",
    "    features_test = df_loan_test_data \n",
    "    \n",
    "    #print(features_test.head())\n",
    " \n",
    "    return loan_numbers, features_test\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Score 90.00 % \n",
      "Training time: 0.573 seconds\n",
      "Prediction time: 0.062 seconds\n"
     ]
    }
   ],
   "source": [
    "#Measure training time\n",
    "t0 = time()\t\n",
    "\n",
    "#Split data\n",
    "features_train, features_test, labels_train, labels_test = pre_process_training_data()\n",
    "\n",
    "#Train model\n",
    "clf = train_model(features_train, labels_train)\n",
    "\n",
    "print (\"Training time: %3.3f seconds\" %  (round(time()-t0, 3)))\n",
    "\n",
    "t0 = time()\t\n",
    "\n",
    "#Load test data\n",
    "loan_numbers, features_test = pre_process_test_data()\n",
    "\n",
    "#Use model for prediction\n",
    "results = clf.predict(features_test)\n",
    "\n",
    "#Create dataframe based on the expected output\n",
    "df_loan_results = pd.DataFrame({'Application_ID' : loan_numbers, 'Loan_Status' : results})\n",
    "\n",
    "#Transform loan status\n",
    "df_loan_results['Loan_Status'] = [ 'Y' if val == 1 else 'N' for val in df_loan_results['Loan_Status']]  \n",
    " \n",
    "#Write the prediction results to output file    \n",
    "df_loan_results.to_csv(\"sample_output.csv\", index=False)\n",
    "\n",
    "print (\"Prediction time: %3.3f seconds\" %  (round(time()-t0, 3)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "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": "3.6.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}