woe

version: 0.1.2

Tools for WoE Transformation mostly used in ScoreCard Model for credit rating

Installation

We can simply use pip to install, as the following:

$ pip install woe

or installing from git

$ pip install git+https://github.com/boredbird/woe

Features

• Split tree with IV criterion
• Rich and plentiful model eval methods
• Unified format and easy for output
• Storage of IV tree for follow-up use

woe aims to only support Python 2.7, so there is no guarantee for Python 3.

woe module function tree

Examples

In the examples directory, there is a simple woe transformation program as tutorials.

Or you can write a more complex program with this woe package.

Version Records

woe 0.1.1 2017-11-28

• woe.config.load_file(): change param data_path to be optional
• woe.eval.eval_feature_stability(): fix bug : psi_dict['stability_index'] computation error
• woe.feature_process.change_feature_dtype(): add friendly tips when encounter a error
• woe.feature_process.calulate_iv(): refactor the code
• woe.feature_process.calculate_iv_split(): refactor the code
• woe.feature_process.binning_data_split(): reduce the number of len() function calls with __len__() and shape attributes;replace namedtuple with dict
• woe.feature_process.fillna(): new added function to fill null value
• woe.GridSearch.grid_search_lr_c(): list of regularization parameter c specified inside the function is changed to the user specified

woe 0.0.9 2017-11-21

• Add module : GridSearch for the search of optimal hyper parametric C in LogisticRegression
• Code refactoring: function compute_ks and plot_ks

woe 0.0.8 2017-09-28

• More flexible: cancel conditional restriction in function feature_process.change_feature_dtype()
• Fix bug: the wrong use of deepcopy in function feature_process.woe_trans()

woe 0.0.7 2017-09-19

• Fix bug: eval.eval_feature_detail raises ValueError('arrays must all be same length')
• Add parameter interface: alpha specified step learning rate ,default 0.01

How to Contribute

Email me,[email protected].