Score card model for Credit Scoring System.
Branch: master
Clone or download
Latest commit 2c77b3d Dec 25, 2017
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.vscode update Nov 12, 2017
ScoreCardModel update Dec 25, 2017
docs Tue Nov 14 16:28:14 2017 Nov 14, 2017
document Mon Dec 11 09:37:10 2017 Dec 11, 2017
requirements mkl:Mon Nov 13 14:14:50 2017 Nov 13, 2017
.gitattributes 🎊 Added .gitattributes & .gitignore files Jun 5, 2017
.gitignore upload Jun 5, 2017
.pmfprc Mon Dec 11 09:37:10 2017 Dec 11, 2017
MANIFEST.in update Oct 31, 2017
README.md Mon Dec 11 09:37:10 2017 Dec 11, 2017
README.rst Mon Dec 11 09:37:10 2017 Dec 11, 2017
example.py
setup.py Mon Dec 11 09:37:10 2017 Dec 11, 2017

README.md

ScoreCardModel

Description

a simple tool for score card model

keywords:math,finance

Feature

  • Serializable
  • mutil classifier model support
  • ks-curve support

Change

  • scorecard now can set a threshold value to return a bool result

Example

>>> from sklearn import datasets
>>> import pandas as pd
>>> from ScoreCardModel.binning.discretization import Discretization
>>> from ScoreCardModel.weight_of_evidence import WeightOfEvidence
>>> from ScoreCardModel.models.logistic_regression_model import LogisticRegressionModel
>>> from ScoreCardModel.score_card import ScoreCardModel
>>>
>>> class MyLR(LogisticRegressionModel):
>>>     def predict(self, x):
>>>          x = self.pre_trade(x)
>>>          return self._predict_proba(x)
>>>      
>>>     def pre_trade(self, x):
>>>         import numpy as np
>>>         result = []
>>>         for i,v in x.items():
>>>             t = self.ds[i].transform([v])[0]
>>>             r = self.woes[i].transform([t])[0]
>>>             result.append(r)
>>>         return np.array(result)
>>>
>>>     def _pre_trade_batch_row(self,row,Y,bins):
>>>         d = Discretization(bins)
>>>         d_row = d.transform(row)
>>>         woe = WeightOfEvidence()
>>>         woe.fit(d_row,Y)
>>>         return d,woe,woe.transform(d_row)
>>>     
>>>     def pre_trade_batch(self, X,Y):
>>>         self.ds = {}
>>>         self.woes = {}
>>>         self.table = {}
>>>         self.ds["sepal length (cm)"],self.woes["sepal length (cm)"],self.table["sepal length (cm)"]= self._pre_trade_batch_row(
>>>             X["sepal length (cm)"],Y,[0,2,5,8])
>>>         self.ds['sepal width (cm)'],self.woes['sepal width (cm)'],self.table['sepal width (cm)'] = self._pre_trade_batch_row(
>>>             X['sepal width (cm)'],Y,[0,2,2.5,3,3.5,5])
>>>         self.ds['petal length (cm)'],self.woes['petal length (cm)'],self.table['petal length (cm)'] = self._pre_trade_batch_row(
>>>             X['petal length (cm)'],Y,[0,1,2,3,4,5,7])
>>>         self.ds['petal width (cm)'],self.woes['petal width (cm)'],self.table['petal width (cm)'] = self._pre_trade_batch_row(
>>>             X['petal width (cm)'],Y,[0,1,2,3])
>>>         return pd.DataFrame(self.table)
>>>
>>> iris = datasets.load_iris()
>>> y = iris.target
>>> z = (y==0)
>>> l = pd.DataFrame(iris.data,columns=iris.feature_names)
>>> lr = MyLR()
>>> lr.train(l,z)
>>> lr.predict(l.loc[0].to_dict())
array([[ 0.46315882,  0.53684118]])
>>> sc = ScoreCardModel(lr)
>>> sc.predict(sc.pre_trade(l.loc[0].to_dict()))
104.3
>>> scs = []
>>> for i in range(len(l)):
>>>    score = sc.predict(sc.pre_trade(l.loc[i].to_dict()))
>>>    scs.append(score)
>>> print(ScoreCardWithKSModel.Threshold_to_score(scs, 0.5))
1.0
>>> print(ScoreCardWithKSModel.Score_to_threshold(scs, score=70))
1.0
             precision    recall  f1-score   support

      False       1.00      1.00      1.00        29
       True       1.00      1.00      1.00        16

avg / total       1.00      1.00      1.00        45
>>> print(ScoreCardWithKSModel.Score_to_threshold(scs, y=z, score=100))
0.3467
>>> print(ScoreCardWithKSModel.Get_ks(scs, y=z, threshold=0.4).ks)
0.9
>>> # ScoreCardWithKSModel.Drawks(scs, y=z)
>>> scsc = [l.loc[i].to_dict() for i in range(len(l))]
>>> scks = ScoreCardWithKSModel.From_scorecard(sc)
>>> print(scks.threshold_to_score(scsc, 0.5))
1.0
>>> print(scks.score_to_threshold(scsc, score=70))
1.0
             precision    recall  f1-score   support

      False       1.00      1.00      1.00        29
       True       1.00      1.00      1.00        16

avg / total       1.00      1.00      1.00        45
>>> print(scks.score_to_threshold(scsc, y=z, score=100))
0.3467
>>> print(scks.get_ks(scsc, y=z, threshold=0.4).ks)
0.9
>>> scks.drawks(scsc, y=z)

Install

python -m pip install ScoreCardModel

Documentation

Documentation on github page https://data-science-tools.github.io/ScoreCardModel/