%matplotlib inline
from pandas import Series, DataFrame
import pandas as pd
import numpy as np
import os
import matplotlib.pylab as plt
from sklearn.cross_validation import train_test_split
from sklearn.linear_model import LassoLarsCV
pd.set_option(’display.float_format’, lambda x:’%.3f’%x)
pd.set_option(’display.mpl_style’, ’default’) # --deprecated
plt.style.use(’ggplot’) # Make the graphs a bit prettier
plt.rcParams[’figure.figsize’] = (15, 5)
Load the dataset
loans = pd.read_csv("./LendingClub.csv", low_memory = False)
LendingClub.csv is a dataset taken from The LendingClub (https://www.lendingclub.com/)
which is a peer-to-peer leading company that directly connects borrowers and potential
lenders/investors
Exploring the target column
_The target column (label column) of the dataset that we are interested in is called
‘bad_loans‘. In this column **1 means a risky (bad) loan 0 means a safe loan.
In order to make this more intuitive, we reassign the target to be:* ** 1 ** as a safe loan,
- ** 0 ** as a risky (bad) loan. We put this in a new column called ‘safe_loans‘**_
loans[’safe_loans’] = loans[’bad_loans’].apply(lambda x : 1 if x == 0 else 0)
loans.drop(’bad_loans’, axis = 1, inplace = True)
Select features to handle
predictors = [’grade’, # grade of the loan
’sub_grade’, # sub-grade of the loan
’short_emp’, # one year or less of employment
’emp_length_num’, # number of years of employment
’home_ownership’, # home_ownership status: own, mortgage or rent
’dti’, # debt to income ratio
’purpose’, # the purpose of the loan
’term’, # the term of the loan
’last_delinq_none’, # has borrower had a delinquincy
’last_major_derog_none’, # has borrower had 90 day or worse rating
’revol_util’, # percent of available credit being used
’total_rec_late_fee’, # total late fees received to day
]
target = ’safe_loans’ # prediction target (y) (+1 means safe, 0 is risky)
Extract the predictors and target columns
loans = loans[predictors + [target]]
Delete rows where any or all of the data are missing
data_clean = loans.dropna()
Convert categorical variables into binary variables
data_clean = pd.get_dummies(data_clean, prefix_sep = ’=’)
(data_clean.describe()).T
count mean std min 25% 50% n
short emp 122607.0 0.123672 0.329208 0.0 0.00 0.00
emp length num 122607.0 6.370256 3.736014 0.0 3.00 6.00
dti 122607.0 15.496888 7.497442 0.0 9.88 15.26
last delinq none 122607.0 0.588115 0.492177 0.0 0.00 1.00
last major derog none 122607.0 0.873906 0.331957 0.0 1.00 1.00
revol util 122607.0 53.716307 25.723881 0.0 34.80 55.70
total rec late fee 122607.0 0.742344 5.363268 0.0 0.00 0.00
safe loans 122607.0 0.811185 0.391363 0.0 1.00 1.00
grade=A 122607.0 0.181996 0.385843 0.0 0.00 0.00
grade=B 122607.0 0.303180 0.459634 0.0 0.00 0.00
grade=C 122607.0 0.244276 0.429659 0.0 0.00 0.00
grade=D 122607.0 0.156394 0.363230 0.0 0.00 0.00
grade=E 122607.0 0.073324 0.260668 0.0 0.00 0.00
grade=F 122607.0 0.032070 0.176187 0.0 0.00 0.00
grade=G 122607.0 0.008760 0.093183 0.0 0.00 0.00
sub grade=A1 122607.0 0.024362 0.154172 0.0 0.00 0.00
sub grade=A2 122607.0 0.027339 0.163071 0.0 0.00 0.00
sub grade=A3 122607.0 0.032258 0.176684 0.0 0.00 0.00
sub grade=A4 122607.0 0.048880 0.215617 0.0 0.00 0.00
sub grade=A5 122607.0 0.049157 0.216197 0.0 0.00 0.00
sub grade=B1 122607.0 0.047607 0.212935 0.0 0.00 0.00
sub grade=B2 122607.0 0.057876 0.233510 0.0 0.00 0.00
sub grade=B3 122607.0 0.073699 0.261281 0.0 0.00 0.00
sub grade=B4 122607.0 0.067525 0.250930 0.0 0.00 0.00
sub grade=B5 122607.0 0.056473 0.230834 0.0 0.00 0.00
sub grade=C1 122607.0 0.057648 0.233077 0.0 0.00 0.00
sub grade=C2 122607.0 0.054858 0.227704 0.0 0.00 0.00
sub grade=C3 122607.0 0.046408 0.210369 0.0 0.00 0.00
sub grade=C4 122607.0 0.044059 0.205228 0.0 0.00 0.00
sub grade=C5 122607.0 0.041303 0.198990 0.0 0.00 0.00
... ... ... ... ... ... ...
sub grade=E4 122607.0 0.012895 0.112821 0.0 0.00 0.00
sub grade=E5 122607.0 0.011092 0.104735 0.0 0.00 0.00
sub grade=F1 122607.0 0.009013 0.094506 0.0 0.00 0.00
sub grade=F2 122607.0 0.007585 0.086763 0.0 0.00 0.00
sub grade=F3 122607.0 0.006280 0.078999 0.0 0.00 0.00
sub grade=F4 122607.0 0.005130 0.071442 0.0 0.00 0.00
sub grade=F5 122607.0 0.004062 0.063603 0.0 0.00 0.00
sub grade=G1 122607.0 0.003018 0.054852 0.0 0.00 0.00
sub grade=G2 122607.0 0.001966 0.044292 0.0 0.00 0.00
sub grade=G3 122607.0 0.001362 0.036881 0.0 0.00 0.00
sub grade=G4 122607.0 0.001240 0.035188 0.0 0.00 0.00
sub grade=G5 122607.0 0.001174 0.034251 0.0 0.00 0.00
home ownership=MORTGAGE 122607.0 0.483170 0.499719 0.0 0.00 0.00
home ownership=OTHER 122607.0 0.001460 0.038182 0.0 0.00 0.00
home ownership=OWN 122607.0 0.081097 0.272984 0.0 0.00 0.00
home ownership=RENT 122607.0 0.434274 0.495663 0.0 0.00 0.00
purpose=car 122607.0 0.019371 0.137825 0.0 0.00 0.00
purpose=credit card 122607.0 0.179843 0.384058 0.0 0.00 0.00
purpose=debt consolidation 122607.0 0.556518 0.496797 0.0 0.00 1.00
purpose=home improvement 122607.0 0.061522 0.240286 0.0 0.00 0.00
purpose=house 122607.0 0.008197 0.090165 0.0 0.00 0.00
purpose=major purchase 122607.0 0.031621 0.174991 0.0 0.00 0.00
purpose=medical 122607.0 0.013107 0.113733 0.0 0.00 0.00
purpose=moving 122607.0 0.009624 0.097630 0.0 0.00 0.00
purpose=other 122607.0 0.074115 0.261959 0.0 0.00 0.00
purpose=small business 122607.0 0.026622 0.160976 0.0 0.00 0.00
purpose=vacation 122607.0 0.007014 0.083457 0.0 0.00 0.00
purpose=wedding 122607.0 0.012446 0.110867 0.0 0.00 0.00
term= 36 months 122607.0 0.797679 0.401732 0.0 1.00 1.00
term= 60 months 122607.0 0.202321 0.401732 0.0 0.00 0.00
75% max
short emp 0.00 1.00
emp length num 11.00 11.00
dti 20.85 39.88
last delinq none 1.00 1.00
last major derog none 1.00 1.00
revol util 74.30 150.70
total rec late fee 0.00 208.82
safe loans 1.00 1.00
grade=A 0.00 1.00
grade=B 1.00 1.00
grade=C 0.00 1.00
grade=D 0.00 1.00
grade=E 0.00 1.00
grade=F 0.00 1.00
grade=G 0.00 1.00
sub grade=A1 0.00 1.00
sub grade=A2 0.00 1.00
sub grade=A3 0.00 1.00
sub grade=A4 0.00 1.00
sub grade=A5 0.00 1.00
sub grade=B1 0.00 1.00
sub grade=B2 0.00 1.00
sub grade=B3 0.00 1.00
sub grade=B4 0.00 1.00
sub grade=B5 0.00 1.00
sub grade=C1 0.00 1.00
sub grade=C2 0.00 1.00
sub grade=C3 0.00 1.00
sub grade=C4 0.00 1.00
sub grade=C5 0.00 1.00
... ... ...
sub grade=E4 0.00 1.00
sub grade=E5 0.00 1.00
sub grade=F1 0.00 1.00
sub grade=F2 0.00 1.00
sub grade=F3 0.00 1.00
sub grade=F4 0.00 1.00
sub grade=F5 0.00 1.00
sub grade=G1 0.00 1.00
sub grade=G2 0.00 1.00
sub grade=G3 0.00 1.00
sub grade=G4 0.00 1.00
sub grade=G5 0.00 1.00
home ownership=MORTGAGE 1.00 1.00
home ownership=OTHER 0.00 1.00
home ownership=OWN 0.00 1.00
home ownership=RENT 1.00 1.00
purpose=car 0.00 1.00
purpose=credit card 0.00 1.00
purpose=debt consolidation 1.00 1.00
purpose=home improvement 0.00 1.00
purpose=house 0.00 1.00
purpose=major purchase 0.00 1.00
purpose=medical 0.00 1.00
purpose=moving 0.00 1.00
purpose=other 0.00 1.00
purpose=small business 0.00 1.00
purpose=vacation 0.00 1.00
purpose=wedding 0.00 1.00
term= 36 months 1.00 1.00
term= 60 months 0.00 1.00
[68 rows x 8 columns]
Extract new features names
features = data_clean.columns.values
features = features[features != target]
Modeling and Prediction
predvar = data_clean[features]
predictors = predvar.copy()
target = data_clean.safe_loans
Standardize predictors to have mean=0 and sd=1
from sklearn import preprocessing
for attr in predictors.columns.values:
predictors[attr] = preprocessing.scale(predictors[attr].astype(’float64’))
Split into training and testing sets
pred_train, pred_test, tar_train, tar_test = train_test_split(predictors, target,
test_size = .4,
random_state = 123)
print(’pred_train.shape’, pred_train.shape)
print(’pred_test.shape’, pred_test.shape)
print(’tar_train.shape’, tar_train.shape)
print(’tar_test.shape’, tar_test.shape)
pred train.shape (73564, 67)
pred test.shape (49043, 67)
tar train.shape (73564,)
tar test.shape (49043,)
Specify the lasso regression model
model = LassoLarsCV(cv = 10, precompute = False).fit(pred_train, tar_train)
Print variable names and regression coefficients
pd.DataFrame([dict(zip(predictors.columns, model.coef_))], index=[’coef’]).T
coef
dti -0.031080
emp length num 0.000000
grade=A 0.036706
grade=B 0.015090
grade=C 0.000000
grade=D -0.014125
grade=E -0.017393
grade=F -0.015787
grade=G -0.009659
home ownership=MORTGAGE 0.010588
home ownership=OTHER -0.001163
home ownership=OWN 0.000000
home ownership=RENT -0.007503
last delinq none -0.008092
last major derog none -0.002272
purpose=car 0.000000
purpose=credit card 0.006216
purpose=debt consolidation 0.000000
purpose=home improvement -0.001184
purpose=house 0.000000
purpose=major purchase 0.000000
purpose=medical -0.003115
purpose=moving -0.000223
purpose=other -0.005091
purpose=small business -0.019715
purpose=vacation -0.000638
purpose=wedding 0.002051
revol util -0.012068
short emp -0.009104
sub grade=A1 0.002509
... ...
sub grade=B4 0.000000
sub grade=B5 -0.001996
sub grade=C1 0.001869
sub grade=C2 0.001481
sub grade=C3 0.000000
sub grade=C4 -0.001447
sub grade=C5 0.000000
sub grade=D1 0.000000
sub grade=D2 0.000000
sub grade=D3 -0.000327
sub grade=D4 -0.003696
sub grade=D5 -0.001762
sub grade=E1 0.004096
sub grade=E2 0.000000
sub grade=E3 -0.001249
sub grade=E4 -0.003977
sub grade=E5 -0.001715
sub grade=F1 0.000000
sub grade=F2 0.000000
sub grade=F3 -0.003955
sub grade=F4 -0.004832
sub grade=F5 -0.005503
sub grade=G1 -0.000474
sub grade=G2 -0.001004
sub grade=G3 0.000000
sub grade=G4 0.004561
sub grade=G5 0.000000
term= 36 months 0.026618
term= 60 months -0.005477
total rec late fee -0.044206
[67 rows x 1 columns]
Plot coefficient progression
m_log_alphas = -np.log10(model.alphas_)
ax = plt.gca()
plt.plot(m_log_alphas, model.coef_path_.T)
plt.axvline(-np.log10(model.alpha_), linestyle = ’--’, color = ’k’,
label = ’alpha CV’)
plt.ylabel(’Regression Coefficients’)
plt.xlabel(’-log(alpha)’)
plt.title(’Regression Coefficients Progression for Lasso Paths’)
<matplotlib.text.Text at 0x17aaab94390>
Plot mean square error for each fold
m_log_alphascv = -np.log10(model.cv_alphas_)
plt.figure()
plt.plot(m_log_alphascv, model.cv_mse_path_, ’:’)
plt.plot(m_log_alphascv, model.cv_mse_path_.mean(axis = -1), ’k’,
label = ’Average across the folds’, linewidth = 2)
plt.axvline(-np.log10(model.alpha_), linestyle = ’--’, color = ’k’,
label = ’alpha CV’)
plt.legend()
plt.xlabel(’-log(alpha)’)
plt.ylabel(’Mean squared error’)
plt.title(’Mean squared error on each fold’)
<matplotlib.text.Text at 0x17aac05cef0>
MSE from training and test data
from sklearn.metrics import mean_squared_error
train_error = mean_squared_error(tar_train, model.predict(pred_train))
test_error = mean_squared_error(tar_test, model.predict(pred_test))
print (’training data MSE’)
print(train_error)
print (’test data MSE’)
print(test_error)
training data MSE
0.141354906717
test data MSE
0.140656085708
R-square from training and test data
rsquared_train = model.score(pred_train, tar_train)
rsquared_test = model.score(pred_test, tar_test)
print (’training data R-square’)
print(rsquared_train)
print (’test data R-square’)
print(rsquared_test)
training data R-square
0.0799940399148
test data R-square
0.0772929635462
Top comments (0)