import pandas as pd

import numpy as np

from sklearn.linear_model import LogisticRegression

from sklearn.cross_validation import KFold #For K-fold cross validation

from sklearn.ensemble import RandomForestClassifier

from sklearn.tree import DecisionTreeClassifier, export_graphviz

from sklearn import metrics

from sklearn.metrics import roc_curve, auc

import ml_metrics as metric

from sklearn import metrics

from sklearn.metrics import roc_curve, auc

df= pd.read_csv('cs-training.csv')

def num_missing(x):

return sum(x.isnull())

df.to_csv('mytraining.csv')

m=df.dropna()

sacuvaj=m.to_csv('deletedNans.csv')

m = m[m['age'] > 0] #Age=0 Deleted

m= m[m['NumberOfTime30-59DaysPastDueNotWorse'] < 96]

m= m[m['NumberOfTime60-89DaysPastDueNotWorse']<96]

m= m[m['NumberOfTimes90DaysLate']<96]

sample_size = len(m[m.SeriousDlqin2yrs == 1]) # get the total count of low-frequency group

small_indices = m[m.SeriousDlqin2yrs == 1].index

small_sample = m.loc[small_indices]

big_indices = m[m.SeriousDlqin2yrs == 0].index

random_indices = np.random.choice(big_indices, sample_size, replace=False) # use the low-frequency group count to randomly sample from high-frequency group

big_sample = m.loc[random_indices]

# Merging all the low-frequency group sample and the new (randomly selected) high-frequency sample together

merged_sample = pd.concat([small_sample, big_sample], ignore_index=True)

#print merged_sample

def classification_model(model, m, predictors, outcome):

model.fit(m[predictors],m[outcome])

#outcome_var = 'SeriousDlqin2yrs'

#model = LogisticRegression()#

#predictor_var = ['RevolvingUtilizationOfUnsecuredLines','age','NumberOfTime30-59DaysPastDueNotWorse','DebtRatio','MonthlyIncome','NumberOfOpenCreditLinesAndLoans','NumberOfTimes90DaysLate','NumberRealEstateLoansOrLines','NumberOfTime60-89DaysPastDueNotWorse','NumberOfDependents']

#classification_model(model, merged_sample,predictor_var,outcome_var)

#outcome_var = 'SeriousDlqin2yrs'

#model = DecisionTreeClassifier(criterion='gini') #Accuracy : 76.991%

#Auc : 77.035%

##Recall : 75.953%#

#Cross-Validation Score : 66.065%

#predictor_var = ['RevolvingUtilizationOfUnsecuredLines','age','NumberOfTime30-59DaysPastDueNotWorse','DebtRatio','MonthlyIncome','NumberOfOpenCreditLinesAndLoans','NumberOfTimes90DaysLate','NumberRealEstateLoansOrLines','NumberOfTime60-89DaysPastDueNotWorse','NumberOfDependents']

#classification_model(model, merged_sample,predictor_var,outcome_var)

#outcome_var = 'SeriousDlqin2yrs'

##model = RandomForestClassifier(max_depth=5)

#predictor_var = ['RevolvingUtilizationOfUnsecuredLines','age','NumberOfTime30-59DaysPastDueNotWorse','DebtRatio','MonthlyIncome','NumberOfOpenCreditLinesAndLoans','NumberOfTimes90DaysLate','NumberRealEstateLoansOrLines','NumberOfTime60-89DaysPastDueNotWorse','NumberOfDependents']

#classification_model(model, merged_sample,predictor_var,outcome_var) #Accuracy : 98.253% #Cross-Validation Score : 70.484%

testfile = pd.read_csv('cs-test.csv')

testfile.info()

deleteMisValues= testfile.fillna(method='ffill')

deleteMisValues.info()

outcome_var = 'SeriousDlqin2yrs'

model = RandomForestClassifier(max_depth=8, min_samples_leaf=9)

predictor_var = ['RevolvingUtilizationOfUnsecuredLines','age','NumberOfTime30-59DaysPastDueNotWorse','DebtRatio','MonthlyIncome','NumberOfOpenCreditLinesAndLoans','NumberOfTimes90DaysLate','NumberRealEstateLoansOrLines','NumberOfTime60-89DaysPastDueNotWorse','NumberOfDependents']

classification_model(model, merged_sample,predictor_var,outcome_var)

predictions = model.predict(deleteMisValues[predictor_var])

print(predictions)

np.savetxt('CRtest4.csv', predictions, delimiter=',')

np.genfromtxt('CRtest4.csv', delimiter=',')