{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## License \n",
    "\n",
    "Copyright 2020 - 2023 Patrick Hall (jphall@gwu.edu)\n",
    "\n",
    "Licensed under the Apache License, Version 2.0 (the \"License\");\n",
    "you may not use this file except in compliance with the License.\n",
    "You may obtain a copy of the License at\n",
    "\n",
    "    http://www.apache.org/licenses/LICENSE-2.0\n",
    "\n",
    "Unless required by applicable law or agreed to in writing, software\n",
    "distributed under the License is distributed on an \"AS IS\" BASIS,\n",
    "WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
    "See the License for the specific language governing permissions and\n",
    "limitations under the License."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**DISCLAIMER:** This notebook is not legal compliance advice."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Lecture 5: Debugging a Machine Learning Model\n",
    "\n",
    "This notebook provides a basic introduction to three practical model debugging techniques that can be applied to machine learning models: residual analysis, sensitivity analysis, and benchmark models. The notebook starts by loading the UCI credit card default dataset and using h2o to train a GBM model to predict credit card defaults. Then, residual analysis is used to discover and debug an issue with the GBM, and the GBM is retrained and improved. Sensitivity analysis is then conducted to test the GBM credit card default model for fairness and stability. The notebook closes by comparing the GBM predictions to a benchmark linear model's predictions.  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Global hyperparameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "SEED = 12345"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Python imports and inits"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/patrickh/Workspace/GWU_rml/env_rml/lib/python3.6/site-packages/sklearn/ensemble/weight_boosting.py:29: DeprecationWarning: numpy.core.umath_tests is an internal NumPy module and should not be imported. It will be removed in a future NumPy release.\n",
      "  from numpy.core.umath_tests import inner1d\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Checking whether there is an H2O instance running at http://localhost:54321 ..... not found.\n",
      "Attempting to start a local H2O server...\n",
      "  Java Version: openjdk version \"1.8.0_252\"; OpenJDK Runtime Environment (build 1.8.0_252-8u252-b09-1~18.04-b09); OpenJDK 64-Bit Server VM (build 25.252-b09, mixed mode)\n",
      "  Starting server from /home/patrickh/Workspace/GWU_rml/env_rml/lib/python3.6/site-packages/h2o/backend/bin/h2o.jar\n",
      "  Ice root: /tmp/tmpfp1yr6sh\n",
      "  JVM stdout: /tmp/tmpfp1yr6sh/h2o_patrickh_started_from_python.out\n",
      "  JVM stderr: /tmp/tmpfp1yr6sh/h2o_patrickh_started_from_python.err\n",
      "  Server is running at http://127.0.0.1:54321\n",
      "Connecting to H2O server at http://127.0.0.1:54321 ... successful.\n",
      "Warning: Your H2O cluster version is too old (9 months and 25 days)! Please download and install the latest version from http://h2o.ai/download/\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div style=\"overflow:auto\"><table style=\"width:50%\"><tr><td>H2O cluster uptime:</td>\n",
       "<td>01 secs</td></tr>\n",
       "<tr><td>H2O cluster timezone:</td>\n",
       "<td>America/New_York</td></tr>\n",
       "<tr><td>H2O data parsing timezone:</td>\n",
       "<td>UTC</td></tr>\n",
       "<tr><td>H2O cluster version:</td>\n",
       "<td>3.26.0.3</td></tr>\n",
       "<tr><td>H2O cluster version age:</td>\n",
       "<td>9 months and 25 days !!!</td></tr>\n",
       "<tr><td>H2O cluster name:</td>\n",
       "<td>H2O_from_python_patrickh_02yx74</td></tr>\n",
       "<tr><td>H2O cluster total nodes:</td>\n",
       "<td>1</td></tr>\n",
       "<tr><td>H2O cluster free memory:</td>\n",
       "<td>1.879 Gb</td></tr>\n",
       "<tr><td>H2O cluster total cores:</td>\n",
       "<td>24</td></tr>\n",
       "<tr><td>H2O cluster allowed cores:</td>\n",
       "<td>24</td></tr>\n",
       "<tr><td>H2O cluster status:</td>\n",
       "<td>accepting new members, healthy</td></tr>\n",
       "<tr><td>H2O connection url:</td>\n",
       "<td>http://127.0.0.1:54321</td></tr>\n",
       "<tr><td>H2O connection proxy:</td>\n",
       "<td>None</td></tr>\n",
       "<tr><td>H2O internal security:</td>\n",
       "<td>False</td></tr>\n",
       "<tr><td>H2O API Extensions:</td>\n",
       "<td>Amazon S3, XGBoost, Algos, AutoML, Core V3, Core V4</td></tr>\n",
       "<tr><td>Python version:</td>\n",
       "<td>3.6.9 final</td></tr></table></div>"
      ],
      "text/plain": [
       "--------------------------  ---------------------------------------------------\n",
       "H2O cluster uptime:         01 secs\n",
       "H2O cluster timezone:       America/New_York\n",
       "H2O data parsing timezone:  UTC\n",
       "H2O cluster version:        3.26.0.3\n",
       "H2O cluster version age:    9 months and 25 days !!!\n",
       "H2O cluster name:           H2O_from_python_patrickh_02yx74\n",
       "H2O cluster total nodes:    1\n",
       "H2O cluster free memory:    1.879 Gb\n",
       "H2O cluster total cores:    24\n",
       "H2O cluster allowed cores:  24\n",
       "H2O cluster status:         accepting new members, healthy\n",
       "H2O connection url:         http://127.0.0.1:54321\n",
       "H2O connection proxy:\n",
       "H2O internal security:      False\n",
       "H2O API Extensions:         Amazon S3, XGBoost, Algos, AutoML, Core V3, Core V4\n",
       "Python version:             3.6.9 final\n",
       "--------------------------  ---------------------------------------------------"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from rmltk import explain, model # simple module for explaining and training models\n",
    "\n",
    "# h2o Python API and specific classes for gbm\n",
    "import h2o \n",
    "from h2o.estimators.gbm import H2OGradientBoostingEstimator\n",
    "import numpy as np   # array, vector, matrix calculations\n",
    "import pandas as pd  # DataFrame handling\n",
    "import shap          # for visualizing Shapley values\n",
    "\n",
    "\n",
    "pd.options.display.max_columns = 999 # enable display of all columns in notebook\n",
    "\n",
    "# plotting functionality\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.lines import Line2D # necessary for custom legends\n",
    "import seaborn as sns\n",
    "\n",
    "# display plots in notebook\n",
    "%matplotlib inline\n",
    "\n",
    "h2o.init(max_mem_size='2G')       # start h2o\n",
    "h2o.remove_all()                  # remove any existing data structures from h2o memory"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Download, Explore, and Prepare UCI Credit Card Default Data\n",
    "\n",
    "UCI credit card default data: https://archive.ics.uci.edu/ml/datasets/default+of+credit+card+clients\n",
    "\n",
    "The UCI credit card default data contains demographic and payment information about credit card customers in Taiwan in the year 2005. The data set contains 23 input variables: \n",
    "\n",
    "* **`LIMIT_BAL`**: Amount of given credit (NT dollar)\n",
    "* **`SEX`**: 1 = male; 2 = female\n",
    "* **`EDUCATION`**: 1 = graduate school; 2 = university; 3 = high school; 4 = others \n",
    "* **`MARRIAGE`**: 1 = married; 2 = single; 3 = others\n",
    "* **`AGE`**: Age in years \n",
    "* **`PAY_0`, `PAY_2` - `PAY_6`**: History of past payment; `PAY_0` = the repayment status in September, 2005; `PAY_2` = the repayment status in August, 2005; ...; `PAY_6` = the repayment status in April, 2005. The measurement scale for the repayment status is: -1 = pay duly; 1 = payment delay for one month; 2 = payment delay for two months; ...; 8 = payment delay for eight months; 9 = payment delay for nine months and above. \n",
    "* **`BILL_AMT1` - `BILL_AMT6`**: Amount of bill statement (NT dollar). `BILL_AMNT1` = amount of bill statement in September, 2005; `BILL_AMT2` = amount of bill statement in August, 2005; ...; `BILL_AMT6` = amount of bill statement in April, 2005. \n",
    "* **`PAY_AMT1` - `PAY_AMT6`**: Amount of previous payment (NT dollar). `PAY_AMT1` = amount paid in September, 2005; `PAY_AMT2` = amount paid in August, 2005; ...; `PAY_AMT6` = amount paid in April, 2005. \n",
    "\n",
    "These 23 input variables are used to predict the target variable, whether or not a customer defaulted on their credit card bill in late 2005.\n",
    "\n",
    "Because h2o accepts both numeric and character inputs, some variables will be recoded into more transparent character values.\n",
    "\n",
    "**Also, best practices and regulations can prohibit the use of demographic variables in credit lending models. This notebook uses them in the model only for educational purposes!**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Import data and clean"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# import XLS file\n",
    "path = 'default_of_credit_card_clients.xls'\n",
    "data = pd.read_excel(path,\n",
    "                     skiprows=1)\n",
    "\n",
    "# remove spaces from target column name \n",
    "data = data.rename(columns={'default payment next month': 'DEFAULT_NEXT_MONTH'}) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Assign modeling roles"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_name = DEFAULT_NEXT_MONTH\n",
      "x_name = ['LIMIT_BAL', 'SEX', 'EDUCATION', 'MARRIAGE', 'AGE', 'PAY_0', 'PAY_2', 'PAY_3', 'PAY_4', 'PAY_5', 'PAY_6', 'BILL_AMT1', 'BILL_AMT2', 'BILL_AMT3', 'BILL_AMT4', 'BILL_AMT5', 'BILL_AMT6', 'PAY_AMT1', 'PAY_AMT2', 'PAY_AMT3', 'PAY_AMT4', 'PAY_AMT5', 'PAY_AMT6']\n"
     ]
    }
   ],
   "source": [
    "# assign target and inputs for GBM\n",
    "y_name = 'DEFAULT_NEXT_MONTH'\n",
    "x_names = [name for name in data.columns if name not in [y_name, 'ID']]\n",
    "print('y_name =', y_name)\n",
    "print('x_name =', x_names)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Helper function for recoding values in the UCI credict card default data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Parse progress: |█████████████████████████████████████████████████████████| 100%\n"
     ]
    }
   ],
   "source": [
    "def recode_cc_data(frame):\n",
    "    \n",
    "    \"\"\" Recodes numeric categorical variables into categorical character variables\n",
    "    with more transparent values. \n",
    "    \n",
    "    Args:\n",
    "        frame: Pandas DataFrame version of UCI credit card default data.\n",
    "        \n",
    "    Returns: \n",
    "        H2OFrame with recoded values.\n",
    "        \n",
    "    \"\"\"\n",
    "    \n",
    "    # define recoded values\n",
    "    sex_dict = {1:'male', 2:'female'}\n",
    "    education_dict = {0:'other', 1:'graduate school', 2:'university', 3:'high school', \n",
    "                      4:'other', 5:'other', 6:'other'}\n",
    "    marriage_dict = {0:'other', 1:'married', 2:'single', 3:'divorced'}\n",
    "    pay_dict = {-2:'no consumption', -1:'pay duly', 0:'use of revolving credit', 1:'1 month delay', \n",
    "                2:'2 month delay', 3:'3 month delay', 4:'4 month delay', 5:'5 month delay', 6:'6 month delay', \n",
    "                7:'7 month delay', 8:'8 month delay', 9:'9+ month delay'}\n",
    "    \n",
    "    # recode values using Pandas apply() and anonymous function\n",
    "    frame['SEX'] = frame['SEX'].apply(lambda i: sex_dict[i])\n",
    "    frame['EDUCATION'] = frame['EDUCATION'].apply(lambda i: education_dict[i])    \n",
    "    frame['MARRIAGE'] = frame['MARRIAGE'].apply(lambda i: marriage_dict[i]) \n",
    "    for name in frame.columns:\n",
    "        if name in ['PAY_0', 'PAY_2', 'PAY_3', 'PAY_4', 'PAY_5', 'PAY_6']:\n",
    "            frame[name] = frame[name].apply(lambda i: pay_dict[i])            \n",
    "                \n",
    "    return h2o.H2OFrame(frame)\n",
    "\n",
    "data = recode_cc_data(data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Ensure target is handled as a categorical variable\n",
    "In h2o, a numeric variable can be treated as numeric or categorical. The target variable `DEFAULT_NEXT_MONTH` takes on values of `0` or `1`. To ensure this numeric variable is treated as a categorical variable, the `asfactor()` function is used to explicitly declare that it is a categorical variable. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "data[y_name] = data[y_name].asfactor() "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Display descriptive statistics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Rows:30000\n",
      "Cols:25\n",
      "\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th>       </th><th>ID               </th><th>LIMIT_BAL         </th><th>SEX   </th><th>EDUCATION      </th><th>MARRIAGE  </th><th>AGE              </th><th>PAY_0                  </th><th>PAY_2                  </th><th>PAY_3                  </th><th>PAY_4                  </th><th>PAY_5                  </th><th>PAY_6                  </th><th>BILL_AMT1        </th><th>BILL_AMT2        </th><th>BILL_AMT3        </th><th>BILL_AMT4        </th><th>BILL_AMT5        </th><th>BILL_AMT6        </th><th>PAY_AMT1          </th><th>PAY_AMT2          </th><th>PAY_AMT3          </th><th>PAY_AMT4          </th><th>PAY_AMT5          </th><th>PAY_AMT6          </th><th>DEFAULT_NEXT_MONTH  </th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td>type   </td><td>int              </td><td>int               </td><td>enum  </td><td>enum           </td><td>enum      </td><td>int              </td><td>enum                   </td><td>enum                   </td><td>enum                   </td><td>enum                   </td><td>enum                   </td><td>enum                   </td><td>int              </td><td>int              </td><td>int              </td><td>int              </td><td>int              </td><td>int              </td><td>int               </td><td>int               </td><td>int               </td><td>int               </td><td>int               </td><td>int               </td><td>enum                </td></tr>\n",
       "<tr><td>mins   </td><td>1.0              </td><td>10000.0           </td><td>      </td><td>               </td><td>          </td><td>21.0             </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>-165580.0        </td><td>-69777.0         </td><td>-157264.0        </td><td>-170000.0        </td><td>-81334.0         </td><td>-339603.0        </td><td>0.0               </td><td>0.0               </td><td>0.0               </td><td>0.0               </td><td>0.0               </td><td>0.0               </td><td>                    </td></tr>\n",
       "<tr><td>mean   </td><td>15000.5          </td><td>167484.32266666688</td><td>      </td><td>               </td><td>          </td><td>35.48549999999994</td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>51223.33090000009</td><td>49179.07516666668</td><td>47013.15479999971</td><td>43262.9489666666 </td><td>40311.40096666653</td><td>38871.76039999991</td><td>5663.580500000014 </td><td>5921.16350000001  </td><td>5225.681500000005 </td><td>4826.076866666661 </td><td>4799.387633333302 </td><td>5215.502566666664 </td><td>                    </td></tr>\n",
       "<tr><td>maxs   </td><td>30000.0          </td><td>1000000.0         </td><td>      </td><td>               </td><td>          </td><td>79.0             </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>964511.0         </td><td>983931.0         </td><td>1664089.0        </td><td>891586.0         </td><td>927171.0         </td><td>961664.0         </td><td>873552.0          </td><td>1684259.0         </td><td>896040.0          </td><td>621000.0          </td><td>426529.0          </td><td>528666.0          </td><td>                    </td></tr>\n",
       "<tr><td>sigma  </td><td>8660.398374208891</td><td>129747.66156720225</td><td>      </td><td>               </td><td>          </td><td>9.21790406809016 </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>73635.86057552959</td><td>71173.76878252836</td><td>69349.38742703681</td><td>64332.85613391641</td><td>60797.1557702648 </td><td>59554.10753674574</td><td>16563.280354025763</td><td>23040.870402057226</td><td>17606.961469803115</td><td>15666.159744031993</td><td>15278.305679144793</td><td>17777.465775435332</td><td>                    </td></tr>\n",
       "<tr><td>zeros  </td><td>0                </td><td>0                 </td><td>      </td><td>               </td><td>          </td><td>0                </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>                       </td><td>2008             </td><td>2506             </td><td>2870             </td><td>3195             </td><td>3506             </td><td>4020             </td><td>5249              </td><td>5396              </td><td>5968              </td><td>6408              </td><td>6703              </td><td>7173              </td><td>                    </td></tr>\n",
       "<tr><td>missing</td><td>0                </td><td>0                 </td><td>0     </td><td>0              </td><td>0         </td><td>0                </td><td>0                      </td><td>0                      </td><td>0                      </td><td>0                      </td><td>0                      </td><td>0                      </td><td>0                </td><td>0                </td><td>0                </td><td>0                </td><td>0                </td><td>0                </td><td>0                 </td><td>0                 </td><td>0                 </td><td>0                 </td><td>0                 </td><td>0                 </td><td>0                   </td></tr>\n",
       "<tr><td>0      </td><td>1.0              </td><td>20000.0           </td><td>female</td><td>university     </td><td>married   </td><td>24.0             </td><td>2 month delay          </td><td>2 month delay          </td><td>pay duly               </td><td>pay duly               </td><td>no consumption         </td><td>no consumption         </td><td>3913.0           </td><td>3102.0           </td><td>689.0            </td><td>0.0              </td><td>0.0              </td><td>0.0              </td><td>0.0               </td><td>689.0             </td><td>0.0               </td><td>0.0               </td><td>0.0               </td><td>0.0               </td><td>1                   </td></tr>\n",
       "<tr><td>1      </td><td>2.0              </td><td>120000.0          </td><td>female</td><td>university     </td><td>single    </td><td>26.0             </td><td>pay duly               </td><td>2 month delay          </td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>2 month delay          </td><td>2682.0           </td><td>1725.0           </td><td>2682.0           </td><td>3272.0           </td><td>3455.0           </td><td>3261.0           </td><td>0.0               </td><td>1000.0            </td><td>1000.0            </td><td>1000.0            </td><td>0.0               </td><td>2000.0            </td><td>1                   </td></tr>\n",
       "<tr><td>2      </td><td>3.0              </td><td>90000.0           </td><td>female</td><td>university     </td><td>single    </td><td>34.0             </td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>29239.0          </td><td>14027.0          </td><td>13559.0          </td><td>14331.0          </td><td>14948.0          </td><td>15549.0          </td><td>1518.0            </td><td>1500.0            </td><td>1000.0            </td><td>1000.0            </td><td>1000.0            </td><td>5000.0            </td><td>0                   </td></tr>\n",
       "<tr><td>3      </td><td>4.0              </td><td>50000.0           </td><td>female</td><td>university     </td><td>married   </td><td>37.0             </td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>46990.0          </td><td>48233.0          </td><td>49291.0          </td><td>28314.0          </td><td>28959.0          </td><td>29547.0          </td><td>2000.0            </td><td>2019.0            </td><td>1200.0            </td><td>1100.0            </td><td>1069.0            </td><td>1000.0            </td><td>0                   </td></tr>\n",
       "<tr><td>4      </td><td>5.0              </td><td>50000.0           </td><td>male  </td><td>university     </td><td>married   </td><td>57.0             </td><td>pay duly               </td><td>use of revolving credit</td><td>pay duly               </td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>8617.0           </td><td>5670.0           </td><td>35835.0          </td><td>20940.0          </td><td>19146.0          </td><td>19131.0          </td><td>2000.0            </td><td>36681.0           </td><td>10000.0           </td><td>9000.0            </td><td>689.0             </td><td>679.0             </td><td>0                   </td></tr>\n",
       "<tr><td>5      </td><td>6.0              </td><td>50000.0           </td><td>male  </td><td>graduate school</td><td>single    </td><td>37.0             </td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>64400.0          </td><td>57069.0          </td><td>57608.0          </td><td>19394.0          </td><td>19619.0          </td><td>20024.0          </td><td>2500.0            </td><td>1815.0            </td><td>657.0             </td><td>1000.0            </td><td>1000.0            </td><td>800.0             </td><td>0                   </td></tr>\n",
       "<tr><td>6      </td><td>7.0              </td><td>500000.0          </td><td>male  </td><td>graduate school</td><td>single    </td><td>29.0             </td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>367965.0         </td><td>412023.0         </td><td>445007.0         </td><td>542653.0         </td><td>483003.0         </td><td>473944.0         </td><td>55000.0           </td><td>40000.0           </td><td>38000.0           </td><td>20239.0           </td><td>13750.0           </td><td>13770.0           </td><td>0                   </td></tr>\n",
       "<tr><td>7      </td><td>8.0              </td><td>100000.0          </td><td>female</td><td>university     </td><td>single    </td><td>23.0             </td><td>use of revolving credit</td><td>pay duly               </td><td>pay duly               </td><td>use of revolving credit</td><td>use of revolving credit</td><td>pay duly               </td><td>11876.0          </td><td>380.0            </td><td>601.0            </td><td>221.0            </td><td>-159.0           </td><td>567.0            </td><td>380.0             </td><td>601.0             </td><td>0.0               </td><td>581.0             </td><td>1687.0            </td><td>1542.0            </td><td>0                   </td></tr>\n",
       "<tr><td>8      </td><td>9.0              </td><td>140000.0          </td><td>female</td><td>high school    </td><td>married   </td><td>28.0             </td><td>use of revolving credit</td><td>use of revolving credit</td><td>2 month delay          </td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>11285.0          </td><td>14096.0          </td><td>12108.0          </td><td>12211.0          </td><td>11793.0          </td><td>3719.0           </td><td>3329.0            </td><td>0.0               </td><td>432.0             </td><td>1000.0            </td><td>1000.0            </td><td>1000.0            </td><td>0                   </td></tr>\n",
       "<tr><td>9      </td><td>10.0             </td><td>20000.0           </td><td>male  </td><td>high school    </td><td>single    </td><td>35.0             </td><td>no consumption         </td><td>no consumption         </td><td>no consumption         </td><td>no consumption         </td><td>pay duly               </td><td>pay duly               </td><td>0.0              </td><td>0.0              </td><td>0.0              </td><td>0.0              </td><td>13007.0          </td><td>13912.0          </td><td>0.0               </td><td>0.0               </td><td>0.0               </td><td>13007.0           </td><td>1122.0            </td><td>0.0               </td><td>0                   </td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Train an H2O GBM classifier"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Split data into training and test sets for early stopping"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train data rows = 21060, columns = 25\n",
      "Validation data rows = 8940, columns = 25\n"
     ]
    }
   ],
   "source": [
    "# split into training and validation\n",
    "train, valid = data.split_frame([0.7], seed=SEED)\n",
    "\n",
    "# summarize split\n",
    "print('Train data rows = %d, columns = %d' % (train.shape[0], train.shape[1]))\n",
    "print('Validation data rows = %d, columns = %d' % (valid.shape[0], valid.shape[1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Train h2o GBM classifier\n",
    "Many tuning parameters must be specified to train a GBM using h2o. Typically a grid search would be performed to identify the best parameters for a given modeling task using the `H2OGridSearch` class. For brevity's sake, a previously-discovered set of good tuning parameters are specified here."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gbm Model Build progress: |███████████████████████████████████████████████| 100%\n",
      "GBM Validation AUC = 0.7804\n"
     ]
    }
   ],
   "source": [
    "# initialize GBM model\n",
    "gbm = H2OGradientBoostingEstimator(ntrees=150,            # maximum 150 trees in GBM\n",
    "                                   max_depth=4,           # trees can have maximum depth of 4\n",
    "                                   sample_rate=0.9,       # use 90% of rows in each iteration (tree)\n",
    "                                   col_sample_rate=0.9,   # use 90% of variables in each iteration (tree)\n",
    "                                   stopping_rounds=5,     # stop if validation error does not decrease for 5 iterations (trees)\n",
    "                                   seed=SEED)            # for reproducibility\n",
    "\n",
    "# train a GBM model\n",
    "gbm.train(y=y_name, x=x_names, training_frame=train, validation_frame=valid)\n",
    "\n",
    "# print AUC\n",
    "print('GBM Validation AUC = %.4f' % gbm.auc(valid=True))\n",
    "\n",
    "# uncomment to see model details\n",
    "# print(model) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Display Shapley variable importance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f82461c5c18>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "shap_contribs_hf = gbm.predict_contributions(valid)\n",
    "shap_contribs_matrix = shap_contribs_hf.as_data_frame().values\n",
    "shap_contribs = shap_contribs_matrix[:,:-1] # remove constant intercept column\n",
    "shap.summary_plot(shap_contribs, x_names, plot_type='bar')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "`PAY_0` is by far the most important variable in the model. Perhaps too important, as will be demonstrated below."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Conduct Residual Analysis to Debug Model\n",
    "Residuals refer to the difference between the recorded value of a dependent variable and the predicted value of a dependent variable for every row in a data set. Plotting the residual values against the predicted values is a time-honored model assessment technique and a great way to see all your modeling results in two dimensions."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Bind model predictions onto validation data \n",
    "To calculate the residuals for our GBM model, first the model predictions are merged onto the validation set. The validation data is used here to see how the model behaves on holdout data, which should be closer to its behavior on new data than analyzing residuals for the training inputs and predictions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gbm prediction progress: |████████████████████████████████████████████████| 100%\n"
     ]
    }
   ],
   "source": [
    "yhat_name = 'p_DEFAULT_NEXT_MONTH'\n",
    "preds1 = gbm.predict(valid).drop(['predict', 'p0'])\n",
    "preds1.columns = [yhat_name]\n",
    "valid_yhat = valid.cbind(preds1[yhat_name])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Calculate logloss residuals for binomial classification\n",
    "Logloss is one measure the GBM model used to make it's decisions. Logloss residuals can be calculated, analyzed, and explained in many ways to understand more about the model's errors. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean logloss residual: 0.431811\n",
      "Logloss from h2o 0.431811\n"
     ]
    }
   ],
   "source": [
    "# shortcut name\n",
    "resid_name = 'r_DEFAULT_NEXT_MONTH' \n",
    "\n",
    "# convert to pandas for easier handling\n",
    "valid_yhat_df = valid_yhat.as_data_frame()\n",
    "\n",
    "# calculate logloss residuals\n",
    "valid_yhat_df[resid_name] = -valid_yhat_df[y_name]*np.log(valid_yhat_df[yhat_name]) -\\\n",
    "                       (1 - valid_yhat_df[y_name])*np.log(1 - valid_yhat_df[yhat_name])   \n",
    "    \n",
    "# check that logloss is calculated correctly\n",
    "print('Mean logloss residual: %.6f' % valid_yhat_df[resid_name].mean())\n",
    "print('Logloss from h2o %.6f' % gbm.logloss(valid=True))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Plot residuals\n",
    "Plotting residuals is a model debugging and diagnostic tool that enables users to see modeling results, and any anomolies, in a single two-dimensional plot. Here the pink points represent customers who defaulted, and the blue points represent customers who did not. A few potential outliers are visible. There appear to be several cases in the validation data with relatively large residuals. Understanding and addressing the factors that cause these outliers could lead to a more acccurate model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fa707af57b8>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    " # initialize figure\n",
    "fig, ax_ = plt.subplots(figsize=(8, 8))         \n",
    "\n",
    "# plot groups with appropriate color\n",
    "color_list = ['royalblue', 'magenta'] \n",
    "c_idx = 0\n",
    "groups = valid_yhat_df.groupby(y_name) # define groups for levels of PAY_0\n",
    "for name, group in groups:\n",
    "    ax_.plot(group.p_DEFAULT_NEXT_MONTH, group.r_DEFAULT_NEXT_MONTH, \n",
    "             label=' '.join([y_name, str(name)]),\n",
    "             marker='o', linestyle='', color=color_list[c_idx], alpha=0.3)\n",
    "    c_idx += 1\n",
    "    \n",
    "# annotate plot\n",
    "_ = plt.xlabel(yhat_name)\n",
    "_ = plt.ylabel(resid_name)\n",
    "_ = ax_.legend(loc=1)\n",
    "_ = plt.title('Global Logloss Residuals')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Some high-magnitude outlying residuals are visible. Who are these customers? Why is the model so wrong about them? And are they somehow exerting undue influence on other predictions? The model could be retrained without these individuals and retested as a potentially remediation strategy. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Sort data by residuals and display data and residuals\n",
    "Printing a table with model inputs, actual target values, and model predictions sorted by residuals is another simple way to analyze residuals. Customers that defaulted, but were predicted not to, are listed at the top of the table below. The next table contains customers who were predicted to default, but then did not. Also notice the jumps in residual values. These are the potential outliers pictured in the residual plot above. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### `DEFAULT_NEXT_MONTH = 1` residuals"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "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>ID</th>\n",
       "      <th>LIMIT_BAL</th>\n",
       "      <th>SEX</th>\n",
       "      <th>EDUCATION</th>\n",
       "      <th>MARRIAGE</th>\n",
       "      <th>AGE</th>\n",
       "      <th>PAY_0</th>\n",
       "      <th>PAY_2</th>\n",
       "      <th>PAY_3</th>\n",
       "      <th>PAY_4</th>\n",
       "      <th>PAY_5</th>\n",
       "      <th>PAY_6</th>\n",
       "      <th>BILL_AMT1</th>\n",
       "      <th>BILL_AMT2</th>\n",
       "      <th>BILL_AMT3</th>\n",
       "      <th>BILL_AMT4</th>\n",
       "      <th>BILL_AMT5</th>\n",
       "      <th>BILL_AMT6</th>\n",
       "      <th>PAY_AMT1</th>\n",
       "      <th>PAY_AMT2</th>\n",
       "      <th>PAY_AMT3</th>\n",
       "      <th>PAY_AMT4</th>\n",
       "      <th>PAY_AMT5</th>\n",
       "      <th>PAY_AMT6</th>\n",
       "      <th>DEFAULT_NEXT_MONTH</th>\n",
       "      <th>p_DEFAULT_NEXT_MONTH</th>\n",
       "      <th>r_DEFAULT_NEXT_MONTH</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2561</td>\n",
       "      <td>310000</td>\n",
       "      <td>female</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>single</td>\n",
       "      <td>32</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>20138</td>\n",
       "      <td>8267</td>\n",
       "      <td>65993</td>\n",
       "      <td>8543</td>\n",
       "      <td>1695</td>\n",
       "      <td>750</td>\n",
       "      <td>8267</td>\n",
       "      <td>66008</td>\n",
       "      <td>8543</td>\n",
       "      <td>1695</td>\n",
       "      <td>750</td>\n",
       "      <td>7350</td>\n",
       "      <td>1</td>\n",
       "      <td>0.045837</td>\n",
       "      <td>3.082662</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>3016</td>\n",
       "      <td>350000</td>\n",
       "      <td>male</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>married</td>\n",
       "      <td>38</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>16459</td>\n",
       "      <td>4120</td>\n",
       "      <td>44164</td>\n",
       "      <td>35233</td>\n",
       "      <td>884</td>\n",
       "      <td>9924</td>\n",
       "      <td>941</td>\n",
       "      <td>44743</td>\n",
       "      <td>0</td>\n",
       "      <td>884</td>\n",
       "      <td>9924</td>\n",
       "      <td>10824</td>\n",
       "      <td>1</td>\n",
       "      <td>0.050230</td>\n",
       "      <td>2.991137</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>11462</td>\n",
       "      <td>210000</td>\n",
       "      <td>female</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>single</td>\n",
       "      <td>46</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>15655</td>\n",
       "      <td>3918</td>\n",
       "      <td>29881</td>\n",
       "      <td>24247</td>\n",
       "      <td>21664</td>\n",
       "      <td>1556</td>\n",
       "      <td>4854</td>\n",
       "      <td>30366</td>\n",
       "      <td>0</td>\n",
       "      <td>433</td>\n",
       "      <td>1556</td>\n",
       "      <td>14047</td>\n",
       "      <td>1</td>\n",
       "      <td>0.050527</td>\n",
       "      <td>2.985238</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>25772</td>\n",
       "      <td>350000</td>\n",
       "      <td>female</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>married</td>\n",
       "      <td>33</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>82964</td>\n",
       "      <td>68532</td>\n",
       "      <td>17926</td>\n",
       "      <td>17966</td>\n",
       "      <td>30741</td>\n",
       "      <td>31088</td>\n",
       "      <td>68940</td>\n",
       "      <td>18018</td>\n",
       "      <td>18058</td>\n",
       "      <td>30897</td>\n",
       "      <td>31244</td>\n",
       "      <td>88461</td>\n",
       "      <td>1</td>\n",
       "      <td>0.051503</td>\n",
       "      <td>2.966110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>6933</td>\n",
       "      <td>500000</td>\n",
       "      <td>male</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>single</td>\n",
       "      <td>37</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>4331</td>\n",
       "      <td>60446</td>\n",
       "      <td>30592</td>\n",
       "      <td>154167</td>\n",
       "      <td>13410</td>\n",
       "      <td>25426</td>\n",
       "      <td>60446</td>\n",
       "      <td>30594</td>\n",
       "      <td>150843</td>\n",
       "      <td>163881</td>\n",
       "      <td>25426</td>\n",
       "      <td>39526</td>\n",
       "      <td>1</td>\n",
       "      <td>0.051717</td>\n",
       "      <td>2.961960</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      ID  LIMIT_BAL     SEX        EDUCATION MARRIAGE  AGE  \\\n",
       "0   2561     310000  female  graduate school   single   32   \n",
       "1   3016     350000    male  graduate school  married   38   \n",
       "2  11462     210000  female  graduate school   single   46   \n",
       "3  25772     350000  female  graduate school  married   33   \n",
       "4   6933     500000    male  graduate school   single   37   \n",
       "\n",
       "                     PAY_0           PAY_2           PAY_3  \\\n",
       "0           no consumption  no consumption  no consumption   \n",
       "1           no consumption  no consumption        pay duly   \n",
       "2                 pay duly        pay duly        pay duly   \n",
       "3  use of revolving credit        pay duly        pay duly   \n",
       "4                 pay duly        pay duly        pay duly   \n",
       "\n",
       "                     PAY_4                    PAY_5           PAY_6  \\\n",
       "0           no consumption           no consumption  no consumption   \n",
       "1  use of revolving credit  use of revolving credit  no consumption   \n",
       "2  use of revolving credit  use of revolving credit        pay duly   \n",
       "3                 pay duly                 pay duly        pay duly   \n",
       "4                 pay duly                 pay duly        pay duly   \n",
       "\n",
       "   BILL_AMT1  BILL_AMT2  BILL_AMT3  BILL_AMT4  BILL_AMT5  BILL_AMT6  PAY_AMT1  \\\n",
       "0      20138       8267      65993       8543       1695        750      8267   \n",
       "1      16459       4120      44164      35233        884       9924       941   \n",
       "2      15655       3918      29881      24247      21664       1556      4854   \n",
       "3      82964      68532      17926      17966      30741      31088     68940   \n",
       "4       4331      60446      30592     154167      13410      25426     60446   \n",
       "\n",
       "   PAY_AMT2  PAY_AMT3  PAY_AMT4  PAY_AMT5  PAY_AMT6  DEFAULT_NEXT_MONTH  \\\n",
       "0     66008      8543      1695       750      7350                   1   \n",
       "1     44743         0       884      9924     10824                   1   \n",
       "2     30366         0       433      1556     14047                   1   \n",
       "3     18018     18058     30897     31244     88461                   1   \n",
       "4     30594    150843    163881     25426     39526                   1   \n",
       "\n",
       "   p_DEFAULT_NEXT_MONTH  r_DEFAULT_NEXT_MONTH  \n",
       "0              0.045837              3.082662  \n",
       "1              0.050230              2.991137  \n",
       "2              0.050527              2.985238  \n",
       "3              0.051503              2.966110  \n",
       "4              0.051717              2.961960  "
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "valid_yhat_df1 = valid_yhat_df[valid_yhat_df[y_name] == 1]\n",
    "valid_yhat_df1 = valid_yhat_df1.sort_values(by='r_DEFAULT_NEXT_MONTH', ascending=False).reset_index(drop=True)\n",
    "valid_yhat_df1.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### `DEFAULT_NEXT_MONTH = 0` residuals"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "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>ID</th>\n",
       "      <th>LIMIT_BAL</th>\n",
       "      <th>SEX</th>\n",
       "      <th>EDUCATION</th>\n",
       "      <th>MARRIAGE</th>\n",
       "      <th>AGE</th>\n",
       "      <th>PAY_0</th>\n",
       "      <th>PAY_2</th>\n",
       "      <th>PAY_3</th>\n",
       "      <th>PAY_4</th>\n",
       "      <th>PAY_5</th>\n",
       "      <th>PAY_6</th>\n",
       "      <th>BILL_AMT1</th>\n",
       "      <th>BILL_AMT2</th>\n",
       "      <th>BILL_AMT3</th>\n",
       "      <th>BILL_AMT4</th>\n",
       "      <th>BILL_AMT5</th>\n",
       "      <th>BILL_AMT6</th>\n",
       "      <th>PAY_AMT1</th>\n",
       "      <th>PAY_AMT2</th>\n",
       "      <th>PAY_AMT3</th>\n",
       "      <th>PAY_AMT4</th>\n",
       "      <th>PAY_AMT5</th>\n",
       "      <th>PAY_AMT6</th>\n",
       "      <th>DEFAULT_NEXT_MONTH</th>\n",
       "      <th>p_DEFAULT_NEXT_MONTH</th>\n",
       "      <th>r_DEFAULT_NEXT_MONTH</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>5916</td>\n",
       "      <td>110000</td>\n",
       "      <td>female</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>married</td>\n",
       "      <td>41</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.886468</td>\n",
       "      <td>2.175667</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>9672</td>\n",
       "      <td>170000</td>\n",
       "      <td>male</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>single</td>\n",
       "      <td>48</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>2400</td>\n",
       "      <td>2400</td>\n",
       "      <td>2400</td>\n",
       "      <td>2400</td>\n",
       "      <td>2400</td>\n",
       "      <td>2400</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.874018</td>\n",
       "      <td>2.071614</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>22725</td>\n",
       "      <td>100000</td>\n",
       "      <td>female</td>\n",
       "      <td>university</td>\n",
       "      <td>married</td>\n",
       "      <td>38</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>750</td>\n",
       "      <td>750</td>\n",
       "      <td>750</td>\n",
       "      <td>750</td>\n",
       "      <td>750</td>\n",
       "      <td>750</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1500</td>\n",
       "      <td>0</td>\n",
       "      <td>0.869051</td>\n",
       "      <td>2.032945</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>19316</td>\n",
       "      <td>110000</td>\n",
       "      <td>female</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>married</td>\n",
       "      <td>41</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>7 month delay</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>150</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.866568</td>\n",
       "      <td>2.014160</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>29505</td>\n",
       "      <td>20000</td>\n",
       "      <td>male</td>\n",
       "      <td>university</td>\n",
       "      <td>married</td>\n",
       "      <td>40</td>\n",
       "      <td>1 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>14829</td>\n",
       "      <td>17267</td>\n",
       "      <td>16706</td>\n",
       "      <td>18694</td>\n",
       "      <td>19049</td>\n",
       "      <td>18459</td>\n",
       "      <td>3000</td>\n",
       "      <td>0</td>\n",
       "      <td>2560</td>\n",
       "      <td>955</td>\n",
       "      <td>0</td>\n",
       "      <td>661</td>\n",
       "      <td>0</td>\n",
       "      <td>0.852781</td>\n",
       "      <td>1.915832</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      ID  LIMIT_BAL     SEX        EDUCATION MARRIAGE  AGE          PAY_0  \\\n",
       "0   5916     110000  female  graduate school  married   41  2 month delay   \n",
       "1   9672     170000    male  graduate school   single   48  2 month delay   \n",
       "2  22725     100000  female       university  married   38  3 month delay   \n",
       "3  19316     110000  female  graduate school  married   41  3 month delay   \n",
       "4  29505      20000    male       university  married   40  1 month delay   \n",
       "\n",
       "           PAY_2          PAY_3          PAY_4          PAY_5          PAY_6  \\\n",
       "0  2 month delay  7 month delay  7 month delay  7 month delay  7 month delay   \n",
       "1  2 month delay  7 month delay  7 month delay  7 month delay  7 month delay   \n",
       "2  2 month delay  2 month delay  3 month delay  3 month delay  3 month delay   \n",
       "3  2 month delay  2 month delay  7 month delay  7 month delay  7 month delay   \n",
       "4  2 month delay  3 month delay  2 month delay  3 month delay  3 month delay   \n",
       "\n",
       "   BILL_AMT1  BILL_AMT2  BILL_AMT3  BILL_AMT4  BILL_AMT5  BILL_AMT6  PAY_AMT1  \\\n",
       "0        150        150        150        150        150        150         0   \n",
       "1       2400       2400       2400       2400       2400       2400         0   \n",
       "2        750        750        750        750        750        750         0   \n",
       "3        150        150        150        150        150        150         0   \n",
       "4      14829      17267      16706      18694      19049      18459      3000   \n",
       "\n",
       "   PAY_AMT2  PAY_AMT3  PAY_AMT4  PAY_AMT5  PAY_AMT6  DEFAULT_NEXT_MONTH  \\\n",
       "0         0         0         0         0         0                   0   \n",
       "1         0         0         0         0         0                   0   \n",
       "2         0         0         0         0      1500                   0   \n",
       "3         0         0         0         0         0                   0   \n",
       "4         0      2560       955         0       661                   0   \n",
       "\n",
       "   p_DEFAULT_NEXT_MONTH  r_DEFAULT_NEXT_MONTH  \n",
       "0              0.886468              2.175667  \n",
       "1              0.874018              2.071614  \n",
       "2              0.869051              2.032945  \n",
       "3              0.866568              2.014160  \n",
       "4              0.852781              1.915832  "
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "valid_yhat_df0 = valid_yhat_df[valid_yhat_df[y_name] == 0]\n",
    "valid_yhat_df0 = valid_yhat_df0.sort_values(by='r_DEFAULT_NEXT_MONTH', ascending=False).reset_index(drop=True)\n",
    "valid_yhat_df0.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This simple analysis has uncovered some of the most difficult customers for the GBM to correctly predict default. Perhaps because of the high importance of the payment variables, `PAY_0`-`PAY_6`, the GBM struggles to correctly predict several cases in which customers made timely recent payments and then suddenly defaulted and those customers that were chronically late making payments but did not default."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Plot residuals by most important input variable \n",
    "Residuals can also be plotted for important input variables to understand how the values of a single input variable affect prediction errors. When plotted by `PAY_0`, the residuals confirm that the GBM is struggling to accurately predict cases where default status is not correlated with recent payment behavior in an obvious way."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fa704a9a9b0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# use Seaborn FacetGrid for convenience\n",
    "# some seaborn configs\n",
    "sns.set(font_scale=0.9)                                         # legible font size\n",
    "sns.set_style('whitegrid', {'axes.grid': False})                # white background, no grid in plots\n",
    "sns.set_palette(sns.color_palette([\"#4169e1\", \"#ff00ff\"]))      # consistent colors\n",
    "\n",
    "# facet grid of residuals by PAY_0 \n",
    "sorted_ = valid_yhat_df.sort_values(by='PAY_0')                 # sort for better layout of by-groups\n",
    "g = sns.FacetGrid(sorted_, col='PAY_0', hue=y_name, col_wrap=4) # init grid\n",
    "_ = g.map(plt.scatter, yhat_name, resid_name, alpha=0.4)        # plot points\n",
    "_ = g.add_legend(bbox_to_anchor=(0.82, 0.2))                    # legend"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Retrain GBM Classifier Based on Results of Residual Analysis\n",
    "Now that an issue has been discovered using residual analysis, can it be resolved? \n",
    "\n",
    "#### Create a variable that contains information about behavior over time\n",
    "One strategy to improve prediction accuracy is to introduce a new variable that summarizes a customer's spending behavior over time to expose any potential financial instability: the standard deviation of a customer's bill amounts over six months."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "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>ID</th>\n",
       "      <th>LIMIT_BAL</th>\n",
       "      <th>SEX</th>\n",
       "      <th>EDUCATION</th>\n",
       "      <th>MARRIAGE</th>\n",
       "      <th>AGE</th>\n",
       "      <th>PAY_0</th>\n",
       "      <th>PAY_2</th>\n",
       "      <th>PAY_3</th>\n",
       "      <th>PAY_4</th>\n",
       "      <th>PAY_5</th>\n",
       "      <th>PAY_6</th>\n",
       "      <th>BILL_AMT1</th>\n",
       "      <th>BILL_AMT2</th>\n",
       "      <th>BILL_AMT3</th>\n",
       "      <th>BILL_AMT4</th>\n",
       "      <th>BILL_AMT5</th>\n",
       "      <th>BILL_AMT6</th>\n",
       "      <th>PAY_AMT1</th>\n",
       "      <th>PAY_AMT2</th>\n",
       "      <th>PAY_AMT3</th>\n",
       "      <th>PAY_AMT4</th>\n",
       "      <th>PAY_AMT5</th>\n",
       "      <th>PAY_AMT6</th>\n",
       "      <th>DEFAULT_NEXT_MONTH</th>\n",
       "      <th>bill_std</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>20000</td>\n",
       "      <td>female</td>\n",
       "      <td>university</td>\n",
       "      <td>married</td>\n",
       "      <td>24</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>3913</td>\n",
       "      <td>3102</td>\n",
       "      <td>689</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>689</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1761.633219</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>120000</td>\n",
       "      <td>female</td>\n",
       "      <td>university</td>\n",
       "      <td>single</td>\n",
       "      <td>26</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>2682</td>\n",
       "      <td>1725</td>\n",
       "      <td>2682</td>\n",
       "      <td>3272</td>\n",
       "      <td>3455</td>\n",
       "      <td>3261</td>\n",
       "      <td>0</td>\n",
       "      <td>1000</td>\n",
       "      <td>1000</td>\n",
       "      <td>1000</td>\n",
       "      <td>0</td>\n",
       "      <td>2000</td>\n",
       "      <td>1</td>\n",
       "      <td>637.967841</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>90000</td>\n",
       "      <td>female</td>\n",
       "      <td>university</td>\n",
       "      <td>single</td>\n",
       "      <td>34</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>29239</td>\n",
       "      <td>14027</td>\n",
       "      <td>13559</td>\n",
       "      <td>14331</td>\n",
       "      <td>14948</td>\n",
       "      <td>15549</td>\n",
       "      <td>1518</td>\n",
       "      <td>1500</td>\n",
       "      <td>1000</td>\n",
       "      <td>1000</td>\n",
       "      <td>1000</td>\n",
       "      <td>5000</td>\n",
       "      <td>0</td>\n",
       "      <td>6064.518593</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   ID  LIMIT_BAL     SEX   EDUCATION MARRIAGE  AGE                    PAY_0  \\\n",
       "0   1      20000  female  university  married   24            2 month delay   \n",
       "1   2     120000  female  university   single   26                 pay duly   \n",
       "2   3      90000  female  university   single   34  use of revolving credit   \n",
       "\n",
       "                     PAY_2                    PAY_3                    PAY_4  \\\n",
       "0            2 month delay                 pay duly                 pay duly   \n",
       "1            2 month delay  use of revolving credit  use of revolving credit   \n",
       "2  use of revolving credit  use of revolving credit  use of revolving credit   \n",
       "\n",
       "                     PAY_5                    PAY_6  BILL_AMT1  BILL_AMT2  \\\n",
       "0           no consumption           no consumption       3913       3102   \n",
       "1  use of revolving credit            2 month delay       2682       1725   \n",
       "2  use of revolving credit  use of revolving credit      29239      14027   \n",
       "\n",
       "   BILL_AMT3  BILL_AMT4  BILL_AMT5  BILL_AMT6  PAY_AMT1  PAY_AMT2  PAY_AMT3  \\\n",
       "0        689          0          0          0         0       689         0   \n",
       "1       2682       3272       3455       3261         0      1000      1000   \n",
       "2      13559      14331      14948      15549      1518      1500      1000   \n",
       "\n",
       "   PAY_AMT4  PAY_AMT5  PAY_AMT6  DEFAULT_NEXT_MONTH     bill_std  \n",
       "0         0         0         0                   1  1761.633219  \n",
       "1      1000         0      2000                   1   637.967841  \n",
       "2      1000      1000      5000                   0  6064.518593  "
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data2 = data.as_data_frame()\n",
    "data2['bill_std'] = data2[['BILL_AMT1', 'BILL_AMT2', 'BILL_AMT3', 'BILL_AMT4', 'BILL_AMT5', 'BILL_AMT6']].std(axis=1)\n",
    "data2.head(n=3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Convert Pandas DataFrame back to H2OFrame for modeling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Parse progress: |█████████████████████████████████████████████████████████| 100%\n"
     ]
    }
   ],
   "source": [
    "data2 = h2o.H2OFrame(data2)                          # convert \n",
    "data2[y_name] = data2[y_name].asfactor()             # ensure target is handled as a categorical variable\n",
    "train2, valid2 = data2.split_frame([0.7], seed=SEED) # split into training and validation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Retrain GBM with new variable"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gbm Model Build progress: |███████████████████████████████████████████████| 100%\n",
      "GBM Validation AUC = 0.7825\n"
     ]
    }
   ],
   "source": [
    "# initialize GBM model\n",
    "gbm2 = H2OGradientBoostingEstimator(ntrees=150,            # maximum 150 trees in GBM\n",
    "                                    max_depth=6,           # trees can have maximum depth of 6\n",
    "                                    sample_rate=0.9,       # use 90% of rows in each iteration (tree)\n",
    "                                    col_sample_rate=0.85,  # use 90% of variables in each iteration (tree)\n",
    "                                    stopping_rounds=5,     # stop if validation error does not decrease for 5 iterations (trees)\n",
    "                                    seed=SEED)             # for reproducibility\n",
    "\n",
    "# retrain GBM model\n",
    "gbm2.train(y=y_name,\n",
    "           x=x_names + ['bill_std'], # add new variable\n",
    "           training_frame=train2, \n",
    "           validation_frame=valid2)\n",
    "\n",
    "# print AUC\n",
    "print('GBM Validation AUC = %.4f' % gbm2.auc(valid=True))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "While there maybe be other more complex variables or a more optimal set of hyperparameters that could lead to further incremental increases in accuracy, more information is needed to achieve meaningful improvement in prediction performance. In particular, a common measure for credit lending, the customers' debt-to-income ratio, for each payment and billing period could be particularly useful. Spikes in debt-to-income ratio, representing loss of income or large increases in debt, would likely be very indicative of a default and would expose the GBM to information not currently available in the UCI credit card default data. Introducing new data could also de-emphasize `PAY_0`, which would likely result in a more stable model as well."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 5. Perform Sensitivity Analysis to Test Model Performance on Unseen Data\n",
    "\n",
    "Sensitivity analysis investigates whether model behavior and outputs remain stable when data is intentionally perturbed or other changes are simulated in data. Beyond traditional assessment practices, sensitivity analysis of machine learning model predictions is perhaps the most important validation technique for machine learning models. Machine learning models can make drastically differing predictions for only minor changes in input variable values. \n",
    "\n",
    "Here sensitivity analysis is used to understand the impact of  changing the most important input variable, `PAY_0`, and the impact of a sociologically sensitive variable, `SEX`, in the model. If the model changes in reasonable and expected ways when important variable values are changed this can enhance trust in the model. If the contribution of potentially sensitive variables, such as those related to gender, race, age, marital status, or disability status, can be shown to have minimal impact on the model, this is an indication of fairness in the model predictions and can also increase overall trust in the model."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Bind new model predictions onto validation data \n",
    "Typically, a productive exercise in model debugging and validation is to investigate customers with very high or low predicted probabilities to determine if their predictions stay within reasonable bounds when important variables are changed. The predictions from the new, more accurate model are merged onto the validation set to find these potentially interesting customers. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gbm prediction progress: |████████████████████████████████████████████████| 100%\n"
     ]
    }
   ],
   "source": [
    "preds2 = gbm2.predict(valid2).drop(['predict', 'p0'])\n",
    "preds2.columns = [yhat_name]\n",
    "valid_yhat2 = valid2.cbind(preds2[yhat_name])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Find some percentiles of yhat in the validation data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{0: 14020,\n",
       " 99: 25054,\n",
       " 10: 1027,\n",
       " 20: 21932,\n",
       " 30: 6331,\n",
       " 40: 24248,\n",
       " 50: 18409,\n",
       " 60: 22652,\n",
       " 70: 19717,\n",
       " 80: 9413,\n",
       " 90: 23311}"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_percentile_dict = explain.get_percentile_dict(yhat_name, valid_yhat2.as_data_frame(), 'ID')\n",
    "pred_percentile_dict"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Display validation data prediction range\n",
    "Unlike some regression models and neural networks that can produce outrageous predictions for changes in input variables, GBM predictions in new data are bounded by the lowest and highest probability leaf nodes in each constiuent decision tree in the trained model. Extreme predictions are typically not an issue for tree models and classification tasks, but it is often a good idea to check that the model predictions cover a full range of useful values in the validation set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Lowest prediction: "
     ]
    },
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th style=\"text-align: right;\">  DEFAULT_NEXT_MONTH</th><th style=\"text-align: right;\">  p_DEFAULT_NEXT_MONTH</th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td style=\"text-align: right;\">                   0</td><td style=\"text-align: right;\">              0.037913</td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Highest prediction: "
     ]
    },
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th style=\"text-align: right;\">  DEFAULT_NEXT_MONTH</th><th style=\"text-align: right;\">  p_DEFAULT_NEXT_MONTH</th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td style=\"text-align: right;\">                   1</td><td style=\"text-align: right;\">               0.90732</td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "print('Lowest prediction:', valid_yhat2[valid_yhat2['ID'] == int(pred_percentile_dict[0])][[y_name, yhat_name]])\n",
    "print('Highest prediction:', valid_yhat2[valid_yhat2['ID'] == int(pred_percentile_dict[99])][[y_name, yhat_name]])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Use trained model to test predictions for interesting situations: customer least likely to default\n",
    "As a starting point for further analysis, sensitivity analysis is performed for the customer least likely to default. This person has a very low probability of defaulting according to the trained GBM."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th style=\"text-align: right;\">   ID</th><th style=\"text-align: right;\">  LIMIT_BAL</th><th>SEX  </th><th>EDUCATION  </th><th>MARRIAGE  </th><th style=\"text-align: right;\">  AGE</th><th>PAY_0                  </th><th>PAY_2                  </th><th>PAY_3                  </th><th>PAY_4                  </th><th>PAY_5                  </th><th>PAY_6                  </th><th style=\"text-align: right;\">  BILL_AMT1</th><th style=\"text-align: right;\">  BILL_AMT2</th><th style=\"text-align: right;\">  BILL_AMT3</th><th style=\"text-align: right;\">  BILL_AMT4</th><th style=\"text-align: right;\">  BILL_AMT5</th><th style=\"text-align: right;\">  BILL_AMT6</th><th style=\"text-align: right;\">  PAY_AMT1</th><th style=\"text-align: right;\">  PAY_AMT2</th><th style=\"text-align: right;\">  PAY_AMT3</th><th style=\"text-align: right;\">  PAY_AMT4</th><th style=\"text-align: right;\">  PAY_AMT5</th><th style=\"text-align: right;\">  PAY_AMT6</th><th style=\"text-align: right;\">  DEFAULT_NEXT_MONTH</th><th style=\"text-align: right;\">  bill_std</th><th style=\"text-align: right;\">  p_DEFAULT_NEXT_MONTH</th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td style=\"text-align: right;\">14020</td><td style=\"text-align: right;\">     500000</td><td>male </td><td>university </td><td>married   </td><td style=\"text-align: right;\">   42</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td style=\"text-align: right;\">     114374</td><td style=\"text-align: right;\">     130999</td><td style=\"text-align: right;\">     153648</td><td style=\"text-align: right;\">     172515</td><td style=\"text-align: right;\">     178758</td><td style=\"text-align: right;\">     192589</td><td style=\"text-align: right;\">     50012</td><td style=\"text-align: right;\">     51343</td><td style=\"text-align: right;\">     70000</td><td style=\"text-align: right;\">     70000</td><td style=\"text-align: right;\">     80000</td><td style=\"text-align: right;\">     50000</td><td style=\"text-align: right;\">                   0</td><td style=\"text-align: right;\">   29949.4</td><td style=\"text-align: right;\">              0.037913</td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": []
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_case = valid_yhat2[valid_yhat2['ID'] == int(pred_percentile_dict[0])]\n",
    "test_case"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Test effect of changing `SEX`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "`SEX` should not have a large impact on predictions. This could indicate unwanted sociological bias in the GBM model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gbm prediction progress: |████████████████████████████████████████████████| 100%\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th style=\"text-align: right;\">   ID</th><th style=\"text-align: right;\">  LIMIT_BAL</th><th>SEX   </th><th>EDUCATION  </th><th>MARRIAGE  </th><th style=\"text-align: right;\">  AGE</th><th>PAY_0                  </th><th>PAY_2                  </th><th>PAY_3                  </th><th>PAY_4                  </th><th>PAY_5                  </th><th>PAY_6                  </th><th style=\"text-align: right;\">  BILL_AMT1</th><th style=\"text-align: right;\">  BILL_AMT2</th><th style=\"text-align: right;\">  BILL_AMT3</th><th style=\"text-align: right;\">  BILL_AMT4</th><th style=\"text-align: right;\">  BILL_AMT5</th><th style=\"text-align: right;\">  BILL_AMT6</th><th style=\"text-align: right;\">  PAY_AMT1</th><th style=\"text-align: right;\">  PAY_AMT2</th><th style=\"text-align: right;\">  PAY_AMT3</th><th style=\"text-align: right;\">  PAY_AMT4</th><th style=\"text-align: right;\">  PAY_AMT5</th><th style=\"text-align: right;\">  PAY_AMT6</th><th style=\"text-align: right;\">  DEFAULT_NEXT_MONTH</th><th style=\"text-align: right;\">  bill_std</th><th style=\"text-align: right;\">  predict</th><th style=\"text-align: right;\">      p0</th><th style=\"text-align: right;\">      p1</th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td style=\"text-align: right;\">14020</td><td style=\"text-align: right;\">     500000</td><td>female</td><td>university </td><td>married   </td><td style=\"text-align: right;\">   42</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td style=\"text-align: right;\">     114374</td><td style=\"text-align: right;\">     130999</td><td style=\"text-align: right;\">     153648</td><td style=\"text-align: right;\">     172515</td><td style=\"text-align: right;\">     178758</td><td style=\"text-align: right;\">     192589</td><td style=\"text-align: right;\">     50012</td><td style=\"text-align: right;\">     51343</td><td style=\"text-align: right;\">     70000</td><td style=\"text-align: right;\">     70000</td><td style=\"text-align: right;\">     80000</td><td style=\"text-align: right;\">     50000</td><td style=\"text-align: right;\">                   0</td><td style=\"text-align: right;\">   29949.4</td><td style=\"text-align: right;\">        0</td><td style=\"text-align: right;\">0.962739</td><td style=\"text-align: right;\">0.037261</td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": []
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_case = valid_yhat2[valid_yhat2['ID'] == int(pred_percentile_dict[0])]\n",
    "test_case = test_case.drop([yhat_name])\n",
    "test_case['SEX'] = 'female'\n",
    "test_case = test_case.cbind(gbm2.predict(test_case))\n",
    "test_case"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As desired, simulating this person as a different sex does not have a large impact on their probability of default."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Test effect of changing `PAY_0`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Variable importance and residual analysis indicates that the value of `PAY_0` can have a strong effect on model predictions. Measuring the change in predicted probability when the value of `PAY_0` is changed from a timely payment to late payment is probably a good test case for prediction stability. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gbm prediction progress: |████████████████████████████████████████████████| 100%\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th style=\"text-align: right;\">   ID</th><th style=\"text-align: right;\">  LIMIT_BAL</th><th>SEX  </th><th>EDUCATION  </th><th>MARRIAGE  </th><th style=\"text-align: right;\">  AGE</th><th>PAY_0        </th><th>PAY_2                  </th><th>PAY_3                  </th><th>PAY_4                  </th><th>PAY_5                  </th><th>PAY_6                  </th><th style=\"text-align: right;\">  BILL_AMT1</th><th style=\"text-align: right;\">  BILL_AMT2</th><th style=\"text-align: right;\">  BILL_AMT3</th><th style=\"text-align: right;\">  BILL_AMT4</th><th style=\"text-align: right;\">  BILL_AMT5</th><th style=\"text-align: right;\">  BILL_AMT6</th><th style=\"text-align: right;\">  PAY_AMT1</th><th style=\"text-align: right;\">  PAY_AMT2</th><th style=\"text-align: right;\">  PAY_AMT3</th><th style=\"text-align: right;\">  PAY_AMT4</th><th style=\"text-align: right;\">  PAY_AMT5</th><th style=\"text-align: right;\">  PAY_AMT6</th><th style=\"text-align: right;\">  DEFAULT_NEXT_MONTH</th><th style=\"text-align: right;\">  bill_std</th><th style=\"text-align: right;\">  predict</th><th style=\"text-align: right;\">      p0</th><th style=\"text-align: right;\">      p1</th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td style=\"text-align: right;\">14020</td><td style=\"text-align: right;\">     500000</td><td>male </td><td>university </td><td>married   </td><td style=\"text-align: right;\">   42</td><td>2 month delay</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td>use of revolving credit</td><td style=\"text-align: right;\">     114374</td><td style=\"text-align: right;\">     130999</td><td style=\"text-align: right;\">     153648</td><td style=\"text-align: right;\">     172515</td><td style=\"text-align: right;\">     178758</td><td style=\"text-align: right;\">     192589</td><td style=\"text-align: right;\">     50012</td><td style=\"text-align: right;\">     51343</td><td style=\"text-align: right;\">     70000</td><td style=\"text-align: right;\">     70000</td><td style=\"text-align: right;\">     80000</td><td style=\"text-align: right;\">     50000</td><td style=\"text-align: right;\">                   0</td><td style=\"text-align: right;\">   29949.4</td><td style=\"text-align: right;\">        1</td><td style=\"text-align: right;\">0.410981</td><td style=\"text-align: right;\">0.589019</td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": []
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_case = valid_yhat2[valid_yhat2['ID'] == int(pred_percentile_dict[0])]\n",
    "test_case = test_case.drop([yhat_name])\n",
    "test_case['PAY_0'] = '2 month delay' \n",
    "test_case = test_case.cbind(gbm2.predict(test_case))\n",
    "test_case"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "When the value is changed to `two month delay` there is a very large increase in predicted probability. Such a marked change related to the value of one variable is problematic for numerous reasons."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Use trained model to test predictions for interesting situations: customer most likely to default\n",
    "Now the same test will be performed on the customer most likely to default. This person has a very high probability of default under the GBM model. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th style=\"text-align: right;\">   ID</th><th style=\"text-align: right;\">  LIMIT_BAL</th><th>SEX  </th><th>EDUCATION      </th><th>MARRIAGE  </th><th style=\"text-align: right;\">  AGE</th><th>PAY_0        </th><th>PAY_2        </th><th>PAY_3        </th><th>PAY_4        </th><th>PAY_5        </th><th>PAY_6        </th><th style=\"text-align: right;\">  BILL_AMT1</th><th style=\"text-align: right;\">  BILL_AMT2</th><th style=\"text-align: right;\">  BILL_AMT3</th><th style=\"text-align: right;\">  BILL_AMT4</th><th style=\"text-align: right;\">  BILL_AMT5</th><th style=\"text-align: right;\">  BILL_AMT6</th><th style=\"text-align: right;\">  PAY_AMT1</th><th style=\"text-align: right;\">  PAY_AMT2</th><th style=\"text-align: right;\">  PAY_AMT3</th><th style=\"text-align: right;\">  PAY_AMT4</th><th style=\"text-align: right;\">  PAY_AMT5</th><th style=\"text-align: right;\">  PAY_AMT6</th><th style=\"text-align: right;\">  DEFAULT_NEXT_MONTH</th><th style=\"text-align: right;\">  bill_std</th><th style=\"text-align: right;\">  p_DEFAULT_NEXT_MONTH</th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td style=\"text-align: right;\">25054</td><td style=\"text-align: right;\">     500000</td><td>male </td><td>graduate school</td><td>married   </td><td style=\"text-align: right;\">   55</td><td>3 month delay</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">                   1</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">               0.90732</td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": []
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_case = valid_yhat2[valid_yhat2['ID'] == int(pred_percentile_dict[99])]\n",
    "test_case"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Test effect of changing `SEX`\n",
    "Changing the value for `SEX` for this customer changes the predicted probability by a relatively small amount."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gbm prediction progress: |████████████████████████████████████████████████| 100%\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th style=\"text-align: right;\">   ID</th><th style=\"text-align: right;\">  LIMIT_BAL</th><th>SEX   </th><th>EDUCATION      </th><th>MARRIAGE  </th><th style=\"text-align: right;\">  AGE</th><th>PAY_0        </th><th>PAY_2        </th><th>PAY_3        </th><th>PAY_4        </th><th>PAY_5        </th><th>PAY_6        </th><th style=\"text-align: right;\">  BILL_AMT1</th><th style=\"text-align: right;\">  BILL_AMT2</th><th style=\"text-align: right;\">  BILL_AMT3</th><th style=\"text-align: right;\">  BILL_AMT4</th><th style=\"text-align: right;\">  BILL_AMT5</th><th style=\"text-align: right;\">  BILL_AMT6</th><th style=\"text-align: right;\">  PAY_AMT1</th><th style=\"text-align: right;\">  PAY_AMT2</th><th style=\"text-align: right;\">  PAY_AMT3</th><th style=\"text-align: right;\">  PAY_AMT4</th><th style=\"text-align: right;\">  PAY_AMT5</th><th style=\"text-align: right;\">  PAY_AMT6</th><th style=\"text-align: right;\">  DEFAULT_NEXT_MONTH</th><th style=\"text-align: right;\">  bill_std</th><th style=\"text-align: right;\">  predict</th><th style=\"text-align: right;\">       p0</th><th style=\"text-align: right;\">      p1</th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td style=\"text-align: right;\">25054</td><td style=\"text-align: right;\">     500000</td><td>female</td><td>graduate school</td><td>married   </td><td style=\"text-align: right;\">   55</td><td>3 month delay</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">                   1</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">        1</td><td style=\"text-align: right;\">0.0919855</td><td style=\"text-align: right;\">0.908015</td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": []
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_case = valid_yhat2[valid_yhat2['ID'] == int(pred_percentile_dict[99])]\n",
    "test_case = test_case.drop([yhat_name])\n",
    "test_case['SEX'] = 'female'\n",
    "test_case = test_case.cbind(gbm2.predict(test_case))\n",
    "test_case"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Test effect of changing `PAY_0`\n",
    "Switching the riskiest customer's value for `PAY_0` to `pay duly` reduces the their chance of default by a lot, a noticable swing in probability but still a higher probability value, notably greater than common lending cutoffs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gbm prediction progress: |████████████████████████████████████████████████| 100%\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<table>\n",
       "<thead>\n",
       "<tr><th style=\"text-align: right;\">   ID</th><th style=\"text-align: right;\">  LIMIT_BAL</th><th>SEX  </th><th>EDUCATION      </th><th>MARRIAGE  </th><th style=\"text-align: right;\">  AGE</th><th>PAY_0   </th><th>PAY_2        </th><th>PAY_3        </th><th>PAY_4        </th><th>PAY_5        </th><th>PAY_6        </th><th style=\"text-align: right;\">  BILL_AMT1</th><th style=\"text-align: right;\">  BILL_AMT2</th><th style=\"text-align: right;\">  BILL_AMT3</th><th style=\"text-align: right;\">  BILL_AMT4</th><th style=\"text-align: right;\">  BILL_AMT5</th><th style=\"text-align: right;\">  BILL_AMT6</th><th style=\"text-align: right;\">  PAY_AMT1</th><th style=\"text-align: right;\">  PAY_AMT2</th><th style=\"text-align: right;\">  PAY_AMT3</th><th style=\"text-align: right;\">  PAY_AMT4</th><th style=\"text-align: right;\">  PAY_AMT5</th><th style=\"text-align: right;\">  PAY_AMT6</th><th style=\"text-align: right;\">  DEFAULT_NEXT_MONTH</th><th style=\"text-align: right;\">  bill_std</th><th style=\"text-align: right;\">  predict</th><th style=\"text-align: right;\">      p0</th><th style=\"text-align: right;\">      p1</th></tr>\n",
       "</thead>\n",
       "<tbody>\n",
       "<tr><td style=\"text-align: right;\">25054</td><td style=\"text-align: right;\">     500000</td><td>male </td><td>graduate school</td><td>married   </td><td style=\"text-align: right;\">   55</td><td>pay duly</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td>2 month delay</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">       4957</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">                   1</td><td style=\"text-align: right;\">         0</td><td style=\"text-align: right;\">        1</td><td style=\"text-align: right;\">0.328214</td><td style=\"text-align: right;\">0.671786</td></tr>\n",
       "</tbody>\n",
       "</table>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": []
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_case = valid_yhat2[valid_yhat2['ID'] == int(pred_percentile_dict[99])]\n",
    "test_case = test_case.drop([yhat_name])\n",
    "test_case['PAY_0'] = 'pay duly' \n",
    "test_case = test_case.cbind(gbm2.predict(test_case))\n",
    "test_case"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "From this small number of boundary test cases, the GBM model appears stable. However, if large swings in predictions occur for sensitive or important variables, practicioners are urged to retrain unstable models without the problematic variables or combinations of variables, which may unfortunately involve some trial and error. Also, four test cases is woefully inadequate for real-world models. Automated sensitivity analysis across many variables, combinations of variables, and for many different rows of data seems more appropriate for mission-critical machine learning."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 6. Benchmark Models\n",
    "Benchmark models are an excellent model debugging tool. They can be used at training time to understand how a new model differs from an established, trusted model. They can also be used at scoring time to understand if a newer or more complex model is giving different predictions from a previously deployed trusted model or simpler model. If a prediction from a new model is too different from a prediction from a trusted model, this could be indicative of potential accuracy, fairness, or security problems."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Train a benchmark GLM \n",
    "Instead of using a pre-existing model, a stable penalized GLM will be used as a benchmark in this notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "glm Grid Build progress: |████████████████████████████████████████████████| 100%\n",
      "Best penalized GLM mean logloss residual: 0.440206\n",
      "At threshold: 0.246137\n"
     ]
    }
   ],
   "source": [
    "best_glm = model.glm_grid(x_names + ['bill_std'], y_name, train2, valid2, SEED)\n",
    "print('Best penalized GLM mean logloss residual: %.6f' % \n",
    "      best_glm.logloss(valid=True))\n",
    "glm_cut = best_glm.F1(valid=True)[0][0] # get GLM cutoff to create decisions\n",
    "print('At threshold: %.6f' % glm_cut)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Find a comparable GBM cutoff"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GBM mean logloss residual: 0.431708\n",
      "At threshold: 0.227593\n"
     ]
    }
   ],
   "source": [
    "print('GBM mean logloss residual: %.6f' % \n",
    "      gbm2.logloss(valid=True))\n",
    "gbm_cut = gbm2.F1(valid=True)[0][0] # get GBM cutoff to create decisions\n",
    "print('At threshold: %.6f' % gbm_cut)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Find rows where GLM benchmark model is right, but GBM is wrong\n",
    "One interesting place to start comparing a more complex model to a simpler model is when the simple model is right and the complex model is wrong."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "glm prediction progress: |████████████████████████████████████████████████| 100%\n"
     ]
    },
    {
     "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>LIMIT_BAL</th>\n",
       "      <th>SEX</th>\n",
       "      <th>EDUCATION</th>\n",
       "      <th>MARRIAGE</th>\n",
       "      <th>AGE</th>\n",
       "      <th>PAY_0</th>\n",
       "      <th>PAY_2</th>\n",
       "      <th>PAY_3</th>\n",
       "      <th>PAY_4</th>\n",
       "      <th>PAY_5</th>\n",
       "      <th>PAY_6</th>\n",
       "      <th>BILL_AMT1</th>\n",
       "      <th>BILL_AMT2</th>\n",
       "      <th>BILL_AMT3</th>\n",
       "      <th>BILL_AMT4</th>\n",
       "      <th>BILL_AMT5</th>\n",
       "      <th>BILL_AMT6</th>\n",
       "      <th>PAY_AMT1</th>\n",
       "      <th>PAY_AMT2</th>\n",
       "      <th>PAY_AMT3</th>\n",
       "      <th>PAY_AMT4</th>\n",
       "      <th>PAY_AMT5</th>\n",
       "      <th>PAY_AMT6</th>\n",
       "      <th>DEFAULT_NEXT_MONTH</th>\n",
       "      <th>p_glm_DEFAULT_NEXT_MONTH</th>\n",
       "      <th>p_gbm_DEFAULT_NEXT_MONTH</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>300000</td>\n",
       "      <td>female</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>married</td>\n",
       "      <td>45</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>291</td>\n",
       "      <td>0</td>\n",
       "      <td>0.138724</td>\n",
       "      <td>0.234757</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>51</th>\n",
       "      <td>360000</td>\n",
       "      <td>male</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>single</td>\n",
       "      <td>29</td>\n",
       "      <td>1 month delay</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>pay duly</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>77</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>77</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.203638</td>\n",
       "      <td>0.302616</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>65</th>\n",
       "      <td>50000</td>\n",
       "      <td>female</td>\n",
       "      <td>high school</td>\n",
       "      <td>single</td>\n",
       "      <td>37</td>\n",
       "      <td>1 month delay</td>\n",
       "      <td>3 month delay</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>52626</td>\n",
       "      <td>51537</td>\n",
       "      <td>49205</td>\n",
       "      <td>30394</td>\n",
       "      <td>30249</td>\n",
       "      <td>29957</td>\n",
       "      <td>0</td>\n",
       "      <td>24</td>\n",
       "      <td>1500</td>\n",
       "      <td>1000</td>\n",
       "      <td>1201</td>\n",
       "      <td>30592</td>\n",
       "      <td>1</td>\n",
       "      <td>0.356709</td>\n",
       "      <td>0.211182</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80</th>\n",
       "      <td>10000</td>\n",
       "      <td>male</td>\n",
       "      <td>high school</td>\n",
       "      <td>single</td>\n",
       "      <td>23</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>use of revolving credit</td>\n",
       "      <td>2 month delay</td>\n",
       "      <td>6974</td>\n",
       "      <td>7838</td>\n",
       "      <td>9002</td>\n",
       "      <td>9182</td>\n",
       "      <td>9729</td>\n",
       "      <td>9411</td>\n",
       "      <td>1134</td>\n",
       "      <td>1298</td>\n",
       "      <td>478</td>\n",
       "      <td>847</td>\n",
       "      <td>0</td>\n",
       "      <td>175</td>\n",
       "      <td>0</td>\n",
       "      <td>0.182870</td>\n",
       "      <td>0.288847</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>92</th>\n",
       "      <td>440000</td>\n",
       "      <td>female</td>\n",
       "      <td>graduate school</td>\n",
       "      <td>single</td>\n",
       "      <td>36</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>no consumption</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>162000</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.045490</td>\n",
       "      <td>0.298846</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    LIMIT_BAL     SEX        EDUCATION MARRIAGE  AGE                    PAY_0  \\\n",
       "22     300000  female  graduate school  married   45                 pay duly   \n",
       "51     360000    male  graduate school   single   29            1 month delay   \n",
       "65      50000  female      high school   single   37            1 month delay   \n",
       "80      10000    male      high school   single   23  use of revolving credit   \n",
       "92     440000  female  graduate school   single   36           no consumption   \n",
       "\n",
       "                      PAY_2                    PAY_3                    PAY_4  \\\n",
       "22                 pay duly                 pay duly                 pay duly   \n",
       "51           no consumption                 pay duly                 pay duly   \n",
       "65            3 month delay            2 month delay  use of revolving credit   \n",
       "80  use of revolving credit  use of revolving credit  use of revolving credit   \n",
       "92           no consumption           no consumption           no consumption   \n",
       "\n",
       "                      PAY_5                    PAY_6  BILL_AMT1  BILL_AMT2  \\\n",
       "22                 pay duly                 pay duly        291        291   \n",
       "51           no consumption           no consumption          0          0   \n",
       "65  use of revolving credit  use of revolving credit      52626      51537   \n",
       "80  use of revolving credit            2 month delay       6974       7838   \n",
       "92           no consumption           no consumption          0          0   \n",
       "\n",
       "    BILL_AMT3  BILL_AMT4  BILL_AMT5  BILL_AMT6  PAY_AMT1  PAY_AMT2  PAY_AMT3  \\\n",
       "22        291        291        291        291       291       291       291   \n",
       "51         77          0          0          0         0        77         0   \n",
       "65      49205      30394      30249      29957         0        24      1500   \n",
       "80       9002       9182       9729       9411      1134      1298       478   \n",
       "92          0          0          0          0         0         0         0   \n",
       "\n",
       "    PAY_AMT4  PAY_AMT5  PAY_AMT6  DEFAULT_NEXT_MONTH  \\\n",
       "22       291       291       291                   0   \n",
       "51         0         0         0                   0   \n",
       "65      1000      1201     30592                   1   \n",
       "80       847         0       175                   0   \n",
       "92         0    162000         0                   0   \n",
       "\n",
       "    p_glm_DEFAULT_NEXT_MONTH  p_gbm_DEFAULT_NEXT_MONTH  \n",
       "22                  0.138724                  0.234757  \n",
       "51                  0.203638                  0.302616  \n",
       "65                  0.356709                  0.211182  \n",
       "80                  0.182870                  0.288847  \n",
       "92                  0.045490                  0.298846  "
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# copy validation data\n",
    "valid_yhat_df2 = valid_yhat2.as_data_frame().copy(deep=True)\n",
    "\n",
    "# create columns for gbm and glm preds\n",
    "valid_yhat_df2.rename(columns={yhat_name: 'p_gbm_DEFAULT_NEXT_MONTH'}, inplace=True)\n",
    "valid_yhat_df2['p_glm_DEFAULT_NEXT_MONTH'] = best_glm.\\\n",
    "    predict(valid2)['p1'].\\\n",
    "    as_data_frame()\n",
    "\n",
    "# create columns for gbm and glm decisions (i.e. apply cutoff)\n",
    "valid_yhat_df2['gbm_DECISION'] = 0\n",
    "valid_yhat_df2.loc[valid_yhat_df2['p_gbm_DEFAULT_NEXT_MONTH'] > gbm_cut,\n",
    "                   'gbm_DECISION'] = 1\n",
    "valid_yhat_df2['glm_DECISION'] = 0\n",
    "valid_yhat_df2.loc[valid_yhat_df2['p_glm_DEFAULT_NEXT_MONTH'] > glm_cut,\n",
    "                   'glm_DECISION'] = 1\n",
    "\n",
    "# create columns for gbm and glm wrong decisions\n",
    "valid_yhat_df2['gbm_WRONG'] = 0\n",
    "valid_yhat_df2.loc[valid_yhat_df2[y_name] != valid_yhat_df2['gbm_DECISION'], 'gbm_WRONG'] = 1\n",
    "valid_yhat_df2['glm_WRONG'] = 0\n",
    "valid_yhat_df2.loc[valid_yhat_df2[y_name] != valid_yhat_df2['glm_DECISION'], 'glm_WRONG'] = 1\n",
    "\n",
    "# create a subset of preds where gbm is wrong, but glm is right\n",
    "gbm_wrong = valid_yhat_df2.loc[(valid_yhat_df2['gbm_WRONG'] == 1) & \n",
    "                               (valid_yhat_df2['glm_WRONG'] == 0)]\n",
    "\n",
    "gbm_wrong[x_names + [y_name, 'p_glm_DEFAULT_NEXT_MONTH', 'p_gbm_DEFAULT_NEXT_MONTH']].head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Plot rows where GLM benchmark model is right, but GBM is wrong"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fa6b33450b8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# custom legend\n",
    "custom = [Line2D([0], [0], color='royalblue', lw=2),\n",
    "          Line2D([0], [0], color='deeppink', lw=2),\n",
    "          Line2D([0], [0], marker='o', color='w',\n",
    "          markerfacecolor='orange', markersize=3)]\n",
    "\n",
    "# init plot\n",
    "fig, ax = plt.subplots(figsize=(7, 5)) \n",
    "\n",
    "# plot sorted actuals\n",
    "# double index reset orders index by sort variable and\n",
    "# brings index into frame for plotting\n",
    "_ = gbm_wrong[[y_name,'p_gbm_DEFAULT_NEXT_MONTH']].\\\n",
    "            sort_values(by='p_gbm_DEFAULT_NEXT_MONTH').\\\n",
    "            reset_index(drop=True).\\\n",
    "            reset_index().\\\n",
    "            plot(kind='scatter', x='index', y=y_name, color='orange', s=3, ax=ax, \n",
    "                 legend=False)\n",
    "\n",
    "# plot sorted gbm and glm preds \n",
    "_ = gbm_wrong[['p_glm_DEFAULT_NEXT_MONTH', 'p_gbm_DEFAULT_NEXT_MONTH']].\\\n",
    "            sort_values(by='p_gbm_DEFAULT_NEXT_MONTH').\\\n",
    "            reset_index(drop=True).\\\n",
    "            plot(ax=ax, legend=False, \n",
    "                 title='Ranked Predictions for Correct GLM and Incorrect GBM')\n",
    "\n",
    "# annotate plot\n",
    "_ = ax.legend(custom, ['p_glm_DEFAULT_NEXT_MONTH', 'p_gbm_DEFAULT_NEXT_MONTH', \n",
    "                       y_name])\n",
    "_ = ax.set_xlabel('Ranked Row Index')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For a range of probabilities between ~0.2 and ~0.6 there exists a group of customers where a GLM model gives more correct predictions than the more complex GBM model. In the plot above, the yellow points represent the known target labels, the pink line is the sorted GBM model predictions, and the blue line is the GLM predictions for the same customers and target labels. For this group of people the GLM is obviously able to better represent some attribute of the customer's demographics or behaviors. Can the differences between this group of people and the rest of the customers be identified and leveraged to make better predictions?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Descriptive statistics for rows where GLM benchmark model is right, but GBM is wrong"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "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>ID</th>\n",
       "      <th>LIMIT_BAL</th>\n",
       "      <th>AGE</th>\n",
       "      <th>BILL_AMT1</th>\n",
       "      <th>BILL_AMT2</th>\n",
       "      <th>BILL_AMT3</th>\n",
       "      <th>BILL_AMT4</th>\n",
       "      <th>BILL_AMT5</th>\n",
       "      <th>BILL_AMT6</th>\n",
       "      <th>PAY_AMT1</th>\n",
       "      <th>PAY_AMT2</th>\n",
       "      <th>PAY_AMT3</th>\n",
       "      <th>PAY_AMT4</th>\n",
       "      <th>PAY_AMT5</th>\n",
       "      <th>PAY_AMT6</th>\n",
       "      <th>DEFAULT_NEXT_MONTH</th>\n",
       "      <th>bill_std</th>\n",
       "      <th>p_gbm_DEFAULT_NEXT_MONTH</th>\n",
       "      <th>p_glm_DEFAULT_NEXT_MONTH</th>\n",
       "      <th>gbm_DECISION</th>\n",
       "      <th>glm_DECISION</th>\n",
       "      <th>gbm_WRONG</th>\n",
       "      <th>glm_WRONG</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.000000</td>\n",
       "      <td>387.0</td>\n",
       "      <td>387.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>14074.307494</td>\n",
       "      <td>158010.335917</td>\n",
       "      <td>36.741602</td>\n",
       "      <td>24316.416021</td>\n",
       "      <td>24700.674419</td>\n",
       "      <td>25600.105943</td>\n",
       "      <td>24657.000000</td>\n",
       "      <td>22654.297158</td>\n",
       "      <td>20100.162791</td>\n",
       "      <td>2776.273902</td>\n",
       "      <td>2930.260982</td>\n",
       "      <td>1822.436693</td>\n",
       "      <td>2021.971576</td>\n",
       "      <td>1840.865633</td>\n",
       "      <td>5164.028424</td>\n",
       "      <td>0.069767</td>\n",
       "      <td>5190.572204</td>\n",
       "      <td>0.277158</td>\n",
       "      <td>0.193844</td>\n",
       "      <td>0.930233</td>\n",
       "      <td>0.069767</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>8475.155425</td>\n",
       "      <td>139110.497314</td>\n",
       "      <td>10.381101</td>\n",
       "      <td>70201.456042</td>\n",
       "      <td>69611.284225</td>\n",
       "      <td>70450.914928</td>\n",
       "      <td>67177.658606</td>\n",
       "      <td>60073.067598</td>\n",
       "      <td>53222.300942</td>\n",
       "      <td>8628.703418</td>\n",
       "      <td>12493.251481</td>\n",
       "      <td>6093.907597</td>\n",
       "      <td>8385.120506</td>\n",
       "      <td>9212.506376</td>\n",
       "      <td>35502.730094</td>\n",
       "      <td>0.255084</td>\n",
       "      <td>15358.965587</td>\n",
       "      <td>0.060774</td>\n",
       "      <td>0.050228</td>\n",
       "      <td>0.255084</td>\n",
       "      <td>0.255084</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>78.000000</td>\n",
       "      <td>10000.000000</td>\n",
       "      <td>21.000000</td>\n",
       "      <td>-1580.000000</td>\n",
       "      <td>-5978.000000</td>\n",
       "      <td>-2400.000000</td>\n",
       "      <td>-1580.000000</td>\n",
       "      <td>-1773.000000</td>\n",
       "      <td>-2303.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.127516</td>\n",
       "      <td>0.025709</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>7060.000000</td>\n",
       "      <td>40000.000000</td>\n",
       "      <td>29.000000</td>\n",
       "      <td>131.500000</td>\n",
       "      <td>190.500000</td>\n",
       "      <td>281.000000</td>\n",
       "      <td>117.000000</td>\n",
       "      <td>102.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>192.360136</td>\n",
       "      <td>0.238541</td>\n",
       "      <td>0.165909</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>13531.000000</td>\n",
       "      <td>130000.000000</td>\n",
       "      <td>35.000000</td>\n",
       "      <td>1727.000000</td>\n",
       "      <td>1516.000000</td>\n",
       "      <td>1300.000000</td>\n",
       "      <td>1430.000000</td>\n",
       "      <td>1072.000000</td>\n",
       "      <td>1060.000000</td>\n",
       "      <td>896.000000</td>\n",
       "      <td>526.000000</td>\n",
       "      <td>318.000000</td>\n",
       "      <td>390.000000</td>\n",
       "      <td>390.000000</td>\n",
       "      <td>390.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1072.610025</td>\n",
       "      <td>0.261236</td>\n",
       "      <td>0.194743</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>20920.500000</td>\n",
       "      <td>235000.000000</td>\n",
       "      <td>44.000000</td>\n",
       "      <td>12399.000000</td>\n",
       "      <td>14646.000000</td>\n",
       "      <td>15993.000000</td>\n",
       "      <td>17049.500000</td>\n",
       "      <td>17728.000000</td>\n",
       "      <td>15733.500000</td>\n",
       "      <td>2284.000000</td>\n",
       "      <td>2000.000000</td>\n",
       "      <td>1273.000000</td>\n",
       "      <td>1108.000000</td>\n",
       "      <td>1000.000000</td>\n",
       "      <td>1500.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2557.592732</td>\n",
       "      <td>0.302298</td>\n",
       "      <td>0.225916</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>29951.000000</td>\n",
       "      <td>680000.000000</td>\n",
       "      <td>67.000000</td>\n",
       "      <td>568532.000000</td>\n",
       "      <td>577681.000000</td>\n",
       "      <td>577957.000000</td>\n",
       "      <td>565669.000000</td>\n",
       "      <td>524315.000000</td>\n",
       "      <td>476846.000000</td>\n",
       "      <td>100082.000000</td>\n",
       "      <td>200000.000000</td>\n",
       "      <td>80167.000000</td>\n",
       "      <td>135000.000000</td>\n",
       "      <td>162000.000000</td>\n",
       "      <td>443001.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>184306.178369</td>\n",
       "      <td>0.529573</td>\n",
       "      <td>0.528843</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                 ID      LIMIT_BAL         AGE      BILL_AMT1      BILL_AMT2  \\\n",
       "count    387.000000     387.000000  387.000000     387.000000     387.000000   \n",
       "mean   14074.307494  158010.335917   36.741602   24316.416021   24700.674419   \n",
       "std     8475.155425  139110.497314   10.381101   70201.456042   69611.284225   \n",
       "min       78.000000   10000.000000   21.000000   -1580.000000   -5978.000000   \n",
       "25%     7060.000000   40000.000000   29.000000     131.500000     190.500000   \n",
       "50%    13531.000000  130000.000000   35.000000    1727.000000    1516.000000   \n",
       "75%    20920.500000  235000.000000   44.000000   12399.000000   14646.000000   \n",
       "max    29951.000000  680000.000000   67.000000  568532.000000  577681.000000   \n",
       "\n",
       "           BILL_AMT3      BILL_AMT4      BILL_AMT5      BILL_AMT6  \\\n",
       "count     387.000000     387.000000     387.000000     387.000000   \n",
       "mean    25600.105943   24657.000000   22654.297158   20100.162791   \n",
       "std     70450.914928   67177.658606   60073.067598   53222.300942   \n",
       "min     -2400.000000   -1580.000000   -1773.000000   -2303.000000   \n",
       "25%       281.000000     117.000000     102.000000       0.000000   \n",
       "50%      1300.000000    1430.000000    1072.000000    1060.000000   \n",
       "75%     15993.000000   17049.500000   17728.000000   15733.500000   \n",
       "max    577957.000000  565669.000000  524315.000000  476846.000000   \n",
       "\n",
       "            PAY_AMT1       PAY_AMT2      PAY_AMT3       PAY_AMT4  \\\n",
       "count     387.000000     387.000000    387.000000     387.000000   \n",
       "mean     2776.273902    2930.260982   1822.436693    2021.971576   \n",
       "std      8628.703418   12493.251481   6093.907597    8385.120506   \n",
       "min         0.000000       0.000000      0.000000       0.000000   \n",
       "25%         0.000000       0.000000      0.000000       0.000000   \n",
       "50%       896.000000     526.000000    318.000000     390.000000   \n",
       "75%      2284.000000    2000.000000   1273.000000    1108.000000   \n",
       "max    100082.000000  200000.000000  80167.000000  135000.000000   \n",
       "\n",
       "            PAY_AMT5       PAY_AMT6  DEFAULT_NEXT_MONTH       bill_std  \\\n",
       "count     387.000000     387.000000          387.000000     387.000000   \n",
       "mean     1840.865633    5164.028424            0.069767    5190.572204   \n",
       "std      9212.506376   35502.730094            0.255084   15358.965587   \n",
       "min         0.000000       0.000000            0.000000       0.000000   \n",
       "25%         0.000000       0.000000            0.000000     192.360136   \n",
       "50%       390.000000     390.000000            0.000000    1072.610025   \n",
       "75%      1000.000000    1500.000000            0.000000    2557.592732   \n",
       "max    162000.000000  443001.000000            1.000000  184306.178369   \n",
       "\n",
       "       p_gbm_DEFAULT_NEXT_MONTH  p_glm_DEFAULT_NEXT_MONTH  gbm_DECISION  \\\n",
       "count                387.000000                387.000000    387.000000   \n",
       "mean                   0.277158                  0.193844      0.930233   \n",
       "std                    0.060774                  0.050228      0.255084   \n",
       "min                    0.127516                  0.025709      0.000000   \n",
       "25%                    0.238541                  0.165909      1.000000   \n",
       "50%                    0.261236                  0.194743      1.000000   \n",
       "75%                    0.302298                  0.225916      1.000000   \n",
       "max                    0.529573                  0.528843      1.000000   \n",
       "\n",
       "       glm_DECISION  gbm_WRONG  glm_WRONG  \n",
       "count    387.000000      387.0      387.0  \n",
       "mean       0.069767        1.0        0.0  \n",
       "std        0.255084        0.0        0.0  \n",
       "min        0.000000        1.0        0.0  \n",
       "25%        0.000000        1.0        0.0  \n",
       "50%        0.000000        1.0        0.0  \n",
       "75%        0.000000        1.0        0.0  \n",
       "max        1.000000        1.0        0.0  "
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gbm_wrong.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If this group of people can be isolated, either by descriptive statistics, or by more sophisticated means, the training process could be adapted to fix these errors or another model could be used at scoring time to create more accurate predictions. Even if a group cannot be isolated, the two different model predictions could potentially be blended in this range of predicted probabilities."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Shutdown H2O"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Are you sure you want to shutdown the H2O instance running at http://localhost:54321 (Y/N)? y\n",
      "H2O session _sid_abf7 closed.\n"
     ]
    }
   ],
   "source": [
    "# be careful, this can erase your work!\n",
    "h2o.cluster().shutdown(prompt=True)"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.10.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}