Practicing clean code principles.

• Task: Refactor churn_notebook.ipynb into production level code. The notebook solves the problem identify credit card customers that are most likely to churn. (Attrition_Flag-target) The data comes from kaggle.

CLIENTNUM	Attrition_Flag	Customer_Age	Gender	Dependent_count	Education_Level	Marital_Status	Income_Category
768805383	Existing Customer	45	M	3	High School	Married	$60K - $80K
818770008	Existing Customer	49	F	5	Graduate	Single	Less than $40K
713982108	Existing Customer	51	M	3	Graduate	Married	$80K - $120K
769911858	Existing Customer	40	F	4	High School	Unknown	Less than $40K
709106358	Existing Customer	40	M	3	Uneducated	Married	$60K - $80K

Sequence diagram of churn_library.py

• At at a high level, the churn_library.py file, produces distributions on age, churn, marital_status, total transtactions, as well as correlation matrix. The categorical features are target encoded. After this, two models are trained, a LogisticRegression(baseline) and RandomForestClassifier(Grid search). Those models are serialized to pickles for later use. Finally, feature importance on the test-set is computed using shap-values.

Features

CATEGORICAL_FEATURES = [
    'Gender',
    'Education_Level',
    'Marital_Status',
    'Income_Category',
    'Card_Category'
]

NUMERICAL_FEATURES = [
    'Customer_Age',
    'Dependent_count', 
    'Months_on_book',
    'Total_Relationship_Count', 
    'Months_Inactive_12_mon',
    'Contacts_Count_12_mon', 
    'Credit_Limit', 
    'Total_Revolving_Bal',
    'Avg_Open_To_Buy', 
    'Total_Amt_Chng_Q4_Q1', 
    'Total_Trans_Amt',
    'Total_Trans_Ct', 
    'Total_Ct_Chng_Q4_Q1', 
    'Avg_Utilization_Ratio'
]

Sample marital status distribution (EDA):

Sample classification report:

Sample feature_importance_sample:

Principles of production ready code.

• Clean and modular code, DRY principle.
• Optimizing code for efficiency.
• Writing documentation.
• Follow PEP8 standards.
• Readable, simple, concise.

Folder structure.

• This folder snapshot is after one execution.

C:.
│   .gitignore
│   churn_library.py
│   churn_notebook.ipynb
│   churn_script_logging_and_tests.py
│   constants.py
│   LICENSE
│   README.md
│   requirements_py3.8.txt
│
├───data
│       bank_data.csv
│
├───images
│   │
│   ├───eda (outputs)
│   │       age_distribution.png
│   │       churn_distribution.png
│   │       correlation_matrix.png
│   │       marital_status_distribution.png
│   │       total_trans_ct.png
│   │
│   ├───feature_importance (outputs)
│   │       feature_importance.png
│   │
│   └───results (outputs)
│           classification_report_logistic_regression.png
│           classification_report_random_forest.png
│
├───logs (outputs)
│       churn_library.log
│
├───models (outputs)
│       LogisticRegression.pkl
│       RandomForest.pkl

• on churn_library.py we have the refactor code.
• on churn_script_logging_and_tests.py we have a test execution with logging (logs\churn_library.log). No testing framework was used.

Installation and use.

• pip install -r .\requirements_py3.8.txt
• You can set a test run from terminal with python churn_script_logging_and_tests.py. This will clean project´s artifacts from images, models and logs folder; then perform a test run similar to the sequence image of above, with the addition of exception handling and logging. The test consist mainly on checking that the expected files are written to file system, e.g. images and models. The test will read parameters of execution from constants.py.
• Once you make sure everything is working, you can run from terminal python churn_library.py path_to_any_csv to run an execution with main.