ML 特征工程

'''
数据集下载链接:https://pan.baidu.com/s/1GiPu5A_9FIFOn7EveLHEzQ  密码:1582
'''
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import warnings
import seaborn as sns
import gc

warnings.simplefilter('ignore')
sns.set()

plt.rcParams['font.sans-serif'] = ['KaiTi']
plt.rcParams['font.serif'] = ['KaiTi']
sns.set_style("darkgrid",{"font.sans-serif":['KaiTi', 'Arial']})

%matplotlib inline

train_id_file = './train_identity.csv'
train_transaction_file = './train_transaction.csv'
test_id_file = './test_identity.csv'
test_transaction_file = './test_transaction.csv'

def reduce_mem_usage(df):
    """ iterate through all the columns of a dataframe and modify the data type
        to reduce memory usage.        
    """
    start_mem = df.memory_usage().sum() / 1024**2
    print('Memory usage of dataframe is {:.2f} MB'.format(start_mem))
    
    for col in df.columns:
        col_type = df[col].dtype
        
        if col_type != object:
            c_min = df[col].min()
            c_max = df[col].max()
            if str(col_type)[:3] == 'int':
                if c_min > np.iinfo(np.int8).min and c_max < np.iinfo(np.int8).max:
                    df[col] = df[col].astype(np.int8)
                elif c_min > np.iinfo(np.int16).min and c_max < np.iinfo(np.int16).max:
                    df[col] = df[col].astype(np.int16)
                elif c_min > np.iinfo(np.int32).min and c_max < np.iinfo(np.int32).max:
                    df[col] = df[col].astype(np.int32)
                elif c_min > np.iinfo(np.int64).min and c_max < np.iinfo(np.int64).max:
                    df[col] = df[col].astype(np.int64)  
            else:
                if c_min > np.finfo(np.float16).min and c_max < np.finfo(np.float16).max:
                    df[col] = df[col].astype(np.float16)
                elif c_min > np.finfo(np.float32).min and c_max < np.finfo(np.float32).max:
                    df[col] = df[col].astype(np.float32)
                else:
                    df[col] = df[col].astype(np.float64)
        else:
            df[col] = df[col].astype('category')

    end_mem = df.memory_usage().sum() / 1024**2
    print('Memory usage after optimization is: {:.2f} MB'.format(end_mem))
    print('Decreased by {:.1f}%'.format(100 * (start_mem - end_mem) / start_mem))
    
    return df

train_id = reduce_mem_usage(pd.read_csv(train_id_file))
train_transaction = reduce_mem_usage(pd.read_csv(train_transaction_file))
test_id = reduce_mem_usage(pd.read_csv(test_id_file))
test_trainsaction = reduce_mem_usage(pd.read_csv(test_transaction_file))

Memory usage of dataframe is 45.12 MB
Memory usage after optimization is: 10.02 MB
Decreased by 77.8%
Memory usage of dataframe is 1775.15 MB
Memory usage after optimization is: 487.16 MB
Decreased by 72.6%
Memory usage of dataframe is 44.39 MB
Memory usage after optimization is: 9.86 MB
Decreased by 77.8%
Memory usage of dataframe is 1519.24 MB
Memory usage after optimization is: 425.24 MB
Decreased by 72.0%

train_transaction['TransactionAmt'] = train_transaction['TransactionAmt'].astype(float)

total = len(train_transaction)

total

590540

total_amt = train_transaction.groupby(['isFraud'])['TransactionAmt'].sum().sum()
plt.figure(figsize=(12, 5))
plt.subplot(121)
plot_tr = sns.countplot(x='isFraud', data=train_transaction)
plot_tr.set_title('Fraud Tran Distribution \n 0: No | 1: is', fontsize=16)
plot_tr.set_xlabel('is Fraud', fontsize=16)
plot_tr.set_ylabel('count', fontsize=16)
for p in plot_tr.patches:
    height = p.get_height()
    plot_tr.text(p.get_x() + p.get_width()/2.,
                height + 3,
                '{:1.2f}%'.format(height/total * 100),
                ha='center',
                fontsize=15)

percent_amt = (train_transaction.groupby(['isFraud'])['TransactionAmt'].sum())
percent_amt = percent_amt.reset_index()
plt.subplot(122)

plot_tr_2 = sns.barplot(x='isFraud', y='TransactionAmt', dodge=True, data=percent_amt)
plot_tr_2.set_title('% Total Amount in Transaction Amt \n 0: No Fraud | 1: Fraud', fontsize=18)
plot_tr_2.set_xlabel('Is Fraud?', fontsize=16)
plot_tr_2.set_ylabel('Total Transaction Amout Scalar', fontsize=16)

for p in plot_tr_2.patches:
    height = p.get_height()
    plot_tr_2.text(p.get_x() + p.get_width()/2.,
                  height + 3, 
                  '{:1.2f}%'.format(height/total_amt * 100),
                  ha='center', fontsize=15)
                  
plt.subplots_adjust(wspace=0.65)
plt.show()

plt.figure(figsize=(16, 12))
plt.suptitle('Tran Values Distribution', fontsize=22)

plt.subplot(221)
sub_plot_1 = sns.distplot(train_transaction[train_transaction['TransactionAmt'] <= 1000]['TransactionAmt'])
sub_plot_1.set_title('小于1000元的消费对比', fontsize=16)
sub_plot_1.set_xlabel('')
sub_plot_1.set_ylabel('probability', fontsize=15)


plt.subplot(222)
sub_plot_2 = sns.distplot(np.log(train_transaction['TransactionAmt']))
sub_plot_2.set_title('正态转换后的消费情况', fontsize=16)
sub_plot_2.set_xlabel('')
sub_plot_2.set_ylabel('probability', fontsize=15)

plt.figure(figsize=(16, 12))

plt.subplot(212)
sub_plot_3 = plt.scatter(range(train_transaction[train_transaction['isFraud'] == 0].shape[0]), 
                        np.sort(train_transaction[train_transaction['isFraud'] == 0]['TransactionAmt'].values),
                        label='NoFraud', alpha=.2)
sub_plot_3 = plt.scatter(range(train_transaction[train_transaction['isFraud'] == 1].shape[0]),
                        np.sort(train_transaction[train_transaction['isFraud'] == 1]['TransactionAmt'].values),
                        label='IsFraud', alpha=.2)
sub_plot_3 = plt.title('ECDF \n 违约用户与正常用户的消费金额对比', fontsize=18)
sub_plot_3 = plt.legend()

plt.figure(figsize=(16, 12))

plt.subplot(321)
sub_plot_4 = plt.scatter(range(train_transaction[train_transaction['isFraud'] == 0].shape[0]),
                        np.sort(train_transaction[train_transaction['isFraud'] == 0]['TransactionAmt'].values),
                        label='NoFraud', alpha=.4)
plt.title('正常用户的消费区间', fontsize=16)
plt.xlabel('Index', fontsize=15)
plt.ylabel('消费金额', fontsize=15)


plt.subplot(322)
sub_plot_5 = plt.scatter(range(train_transaction[train_transaction['isFraud'] == 1].shape[0]),
                        np.sort(train_transaction[train_transaction['isFraud'] == 1]['TransactionAmt'].values),
                        label='IsFraud', alpha=.2)
plt.title('违约用户的消费区间', fontsize=16)
plt.xlabel('Index', fontsize=15)
plt.ylabel('消费金额', fontsize=15)


plt.show()

tmp = pd.crosstab(train_transaction['ProductCD'], train_transaction['isFraud'], normalize='index')*100
tmp = tmp.reset_index()
tmp.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)

plt.figure(figsize=(14, 10))
plt.suptitle('产品类型展示', fontsize=22)

plt.subplot(221)
plot_1 = sns.countplot(x='ProductCD', data=train_transaction, order=['W', 'H', 'C', 'S', 'R'])


plot_1.set_title('产品代码展示', fontsize=18)
plot_1.set_xlabel('产品代码', fontsize=16)
plot_1.set_ylabel('消费次数', fontsize=17)
plot_1.set_ylim(0, 500000)
for p in plot_1.patches:
    height = p.get_height()
    plot_1.text(p.get_x()+p.get_width()/2.,
               height + 3,
               '{:1.2f}%'.format(height/total * 100),
               ha='center', fontsize=14)
    
plt.subplot(222)
plot_2 = sns.countplot(x='ProductCD', hue='isFraud', data=train_transaction, order=['W', 'H', 'C', 'S', 'R'])
plt.legend(title='Fraud', loc='best', labels=['No', 'Yes'])
plot_2_2 = plot_2.twinx()
plot_2_2 = sns.pointplot(x='ProductCD', y='IsFraud', data=tmp, color='black', order=['W', 'H', 'C', 'S', 'R'], legend=False)
plot_2.set_title('单产品违约消费者对比', fontsize=16)
plot_2.set_xlabel('产品代码', fontsize=16)
plot_2.set_ylabel('消费次数', fontsize=16)

plt.subplot(212)
plot_3 = sns.boxplot(x='ProductCD', y='TransactionAmt', hue='isFraud',
                      data=train_transaction[train_transaction['TransactionAmt'] <= 2000],
                    order=['W', 'H', 'C', 'S', 'R'])
plot_3.set_title('单种产品消费金额', fontsize=16)
plot_3.set_xlabel('产品代码', fontsize=16)
plot_3.set_ylabel('消费金额', fontsize=16)

plt.subplots_adjust(hspace = 0.6, top = 0.85)

plt.show()

train_transaction.loc[train_transaction.card3.isin(train_transaction.card3.value_counts()[train_transaction.card3.value_counts()<200].index), 
                       'card3'] = 'Others'
train_transaction.loc[train_transaction.card5.isin(train_transaction.card5.value_counts()[train_transaction.card5.value_counts()<200].index), 
                      'card5'] = 'Others'

tmp = pd.crosstab(train_transaction['card3'], train_transaction['isFraud'], normalize='index') * 100
tmp = tmp.reset_index()
tmp.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)

tmp2 = pd.crosstab(train_transaction['card5'], train_transaction['isFraud'], normalize='index') * 100
tmp2 = tmp2.reset_index()
tmp2.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)

plt.figure(figsize=(14,22))

plt.subplot(411)
plot_1 = sns.distplot(train_transaction[train_transaction['isFraud'] == 1]['card1'], label='Fraud')
plot_1 = sns.distplot(train_transaction[train_transaction['isFraud'] == 0]['card1'], label='NoFraud')
plot_1.legend()
plot_1.set_title('Card1正常用户与违约用户对比', fontsize=20)
plot_1.set_xlabel('Card1值', fontsize=18)
plot_1.set_ylabel('可能性', fontsize=18)

plt.subplot(412)
plot_2 = sns.distplot(train_transaction[train_transaction['isFraud'] == 1]['card2'].dropna(), label='Fraud')
plot_2 = sns.distplot(train_transaction[train_transaction['isFraud'] == 0]['card2'].dropna(), label='NoFraud')
plot_2.legend()
plot_2.set_title('Card2正常用户与异常用户对比', fontsize=20)
plot_2.set_xlabel('Card2值', fontsize=18)
plot_2.set_ylabel('可能性', fontsize=18)

plt.subplot(413)
plot_3 = sns.countplot(x='card3', data=train_transaction, order=list(tmp.card3.values))
plot_3_2 = plot_3.twinx()
plot_3_2 = sns.pointplot(x='card3', y='IsFraud', data=tmp,
                        color='black', order=list(tmp.card3.values))
plot_3_2.set_ylabel('异常占比', fontsize=16)
plot_3.set_title('Card3正常用户与异常用户对比', fontsize=20)
plot_3.set_xlabel('Card3值', fontsize=18)
plot_3.set_ylabel('消费总笔数', fontsize=18)

for p in plot_3.patches:
    height = p.get_height()
    plot_3.text(p.get_x()+p.get_width()/2.,
               height+25,
               '{:1.2f}%'.format(height/total*100),
               ha='center')
    

plt.subplot(414)
plot_4 = sns.countplot(x='card5', data=train_transaction, order=list(tmp2.card5.values))
plot_4_2 = plot_4.twinx()
plot_4_2 = sns.pointplot(x='card5', y='IsFraud', data=tmp2,
                        color='black', order=list(tmp2.card5.values))
plot_4_2.set_ylabel('异常占比', fontsize=16)
plot_4.set_title('Card5正常用户与异常用户对比', fontsize=20)
plot_4.set_xlabel('Card5值', fontsize=18)
plot_4.set_ylabel('消费总笔数', fontsize=18)

for p in plot_4.patches:
    height = p.get_height()
    plot_4.text(p.get_x()+p.get_width()/2.,
               height+25,
               '{:1.2f}%'.format(height/total*100),
               ha='center')    


plt.subplots_adjust(hspace=0.6, top=0.85)
plt.show()

tmp = pd.crosstab(train_transaction['card4'], train_transaction['isFraud'], normalize='index') * 100
tmp = tmp.reset_index()
tmp.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)

plt.figure(figsize=(14,10))
plt.suptitle('Card4')

plt.subplot(221)
plot_1 = sns.countplot(x='card4', data=train_transaction, 
                       order=['american express', 'discover', 'mastercard', 'visa'])
plot_1.set_title('Card4种类以及数量', fontsize=20)
plot_1.set_xlabel('Card4种类')
plot_1.set_ylabel('总量')
for p in plot_1.patches:
    height = p.get_height()
    plot_1.text(p.get_x() + p.get_width()/2.,
               height + 3,
               '{:1.2f}%'.format(height/total*100),
               ha='center', fontsize=14)
    

plt.subplot(222)
plot_2 = sns.countplot(x='card4', hue='isFraud', data=train_transaction, 
                      order=['american express', 'discover', 'mastercard', 'visa'])
plot_2_2 = plot_2.twinx()
plot_2_2 = sns.pointplot(x='card4', y='IsFraud', data=tmp, color='black', legend=False, 
                         order=['american express', 'discover', 'mastercard', 'visa'])
plot_2_2.set_ylabel('违约占比')
plot_2.set_title('card4正常与违约用户对比', fontsize=20)
plot_2.set_xlabel('Card4种类')
plot_2.set_ylabel('总量')

plt.subplot(212)
plot_3 = sns.boxplot(x='card4', y='TransactionAmt', hue='isFraud', 
                     data=train_transaction[train_transaction['TransactionAmt'] <= 2000])

plot_3.set_title('Card4按产品和目标分配', fontsize=20)
plot_3.set_xlabel('Card4种类')
plot_3.set_ylabel('交易数值')

plt.subplots_adjust(hspace=0.6, top=0.85)

plt.show()

tmp = pd.crosstab(train_transaction['card6'], train_transaction['isFraud'], normalize='index') * 100
tmp = tmp.reset_index()
tmp.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)

plt.figure(figsize=(16, 10))
plt.suptitle('Card6', fontsize=20)

plt.subplot(221)
plot_1 = sns.countplot(x='card6', data=train_transaction)
plot_1.set_title('Card6各种类占比', fontsize=18)
plot_1.set_xlabel('Card6种类', fontsize=16)
plot_1.set_ylabel('总量', fontsize=16)
for p in plot_1.patches:
    height = p.get_height()
    plot_1.text(p.get_x() + p.get_width()/2,
               height + 3,
               '{:1.2f}%'.format(height/total*100),
               ha='center',fontsize=14)

plt.subplot(222)
plot_2 = sns.countplot(x='card6', hue='isFraud', data=train_transaction, order=list(tmp.card6.values))
plot_2_2 = plot_2.twinx()
plot_2_2 = sns.pointplot(x='card6', y='IsFraud', data=tmp, order=['charge card', 'credit', 'debit', 'debit or credit'],
                        color='black', legend=False)
plot_2_2.set_ylim(0,20)
plot_2_2.set_ylabel('Card6各种类异常占比', fontsize=15)
plot_2.set_title('Card6正常用户与异常用户占比', fontsize=18)
plot_2.set_xlabel('Card6种类', fontsize=15)
plot_2.set_ylabel('总量', fontsize=15)

plt.subplot(212)
plot_3 = sns.boxplot(x='card6', y='TransactionAmt', hue='isFraud', 
                     data=train_transaction[train_transaction['TransactionAmt'] <= 2000])

plot_3.set_title('Card6正常用户与异常用户交易金额对比', fontsize=18)
plot_3.set_xlabel('Card6种类', fontsize=15)
plot_3.set_ylabel('交易金额', fontsize=15)


plt.subplots_adjust(hspace=0.6, top=0.85)
plt.show()

for Columns in ['M1', 'M2', 'M3', 'M4', 'M5', 'M6', 'M7', 'M8', 'M9']:
    if 'Miss' not in train_transaction[Columns].cat.categories:
        train_transaction[Columns] = train_transaction[Columns].cat.add_categories(['Miss']);
    train_transaction[Columns] = train_transaction[Columns].fillna('Miss')

    
def ploting_dist_ratio(DataFile, Column, lim=2000):
    tmp = pd.crosstab(train_transaction[Column], train_transaction['isFraud'], normalize='index') * 100
    tmp = tmp.reset_index()
    tmp.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)

    plt.figure(figsize=(20,5))
    plt.suptitle(f'{Column}', fontsize=20)

    plt.subplot(121)
    plot_1 = sns.countplot(x=Column, data=DataFile, order=list(tmp[Column].values))
    plot_1.set_title(f'{Column}的常规展示以及异常占比展示', fontsize=18)
    plot_1.set_ylim(0, 400000)
        
    
    plot_1_1 = plot_1.twinx()
    plot_1_1 = sns.pointplot(x=Column, y='IsFraud', data=tmp, order=list(tmp[Column].values), 
                             color='black', legend=False)
    plot_1_1.set_ylim(0,20)
    plot_1_1.set_ylabel('异常占比', fontsize=16)
    plot_1.set_xlabel(f'{Column} 种类')
    plot_1.set_ylabel('总量', fontsize=16)
    
    for p in plot_1.patches:
        height = p.get_height()
        plot_1.text(p.get_x() + p.get_width()/2,
                   height + 3,'{:1.2f}%'.format(height/total*100),
                   ha='center', fontsize=14)
    
    
    perc_amt = (train_transaction.groupby(['isFraud', Column])['TransactionAmt'].sum() / total_amt * 100).unstack('isFraud')
    print(perc_amt)
    perc_amt = perc_amt.reset_index()
    perc_amt.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)
    
   
    
    plt.subplot(122)
    plot_2 = sns.boxplot(x=Column , y='TransactionAmt', hue='isFraud', data=DataFile[DataFile['TransactionAmt'] <= lim],
                         order=list(tmp[Column].values))
    plot_2_2 = plot_2.twinx()
    plot_2_2 = sns.pointplot(x=Column, y='IsFraud', data=perc_amt, order=list(tmp[Column].values),
                        color='black', legend=False)
    plot_2_2.set_ylim(0,5)
    plot_2_2.set_ylabel('欺诈总额', fontsize=16)
    plot_2.set_title(f'{Column} 种类消费金额')
    plot_2.set_xlabel(f'{Column} 种类')
    plot_2.set_ylabel('金额 (U$)')
    
    
    plt.subplots_adjust(hspace=0.65, top=0.85, wspace=0.35)
    plt.show()

for Columns in ['M1', 'M2', 'M3', 'M4', 'M5', 'M6', 'M7', 'M8', 'M9']:
    ploting_dist_ratio(train_transaction, Columns)

isFraud          0         1
M1                          
F         0.004843       NaN
T        59.554396  1.898156
Miss     36.573351  1.969254

isFraud          0         1
M2                          
F         6.912159  0.342055
T        52.647080  1.556101
Miss     36.573351  1.969254

isFraud          0         1
M3                          
F        13.725183  0.554626
T        45.834056  1.343530
Miss     36.573351  1.969254

isFraud          0         1
M4                          
M0       31.412624  1.958153
M1        8.834237  0.280888
M2        3.381058  0.449007
Miss     52.504671  1.179362

isFraud          0         1
M5                          
F        24.187301  1.255668
T        15.947904  0.945038
Miss     55.997385  1.666704

isFraud          0         1
M6                          
F        38.162428  1.351420
T        40.035646  1.146065
Miss     17.934516  1.369926

isFraud          0         1
M7                          
F        41.357407  1.295796
T         6.205311  0.270727
Miss     48.569872  2.300887

isFraud          0         1
M8                          
F        31.564724  1.096196
T        16.006749  0.470327
Miss     48.561117  2.300887

isFraud          0         1
M9                          
F         7.812824  0.366394
T        39.758649  1.200130
Miss     48.561117  2.300887

train_transaction.groupby(['isFraud', 'M1'])['TransactionAmt'].sum()

isFraud  M1  
0        F       3.862156e+03
         T       4.748811e+07
         Miss    2.916324e+07
1        T       1.513572e+06
         Miss    1.570264e+06
Name: TransactionAmt, dtype: float64

train_transaction.loc[train_transaction.\
                      addr1.isin(train_transaction.addr1.\
                                 value_counts()[train_transaction.addr1.\
                                                value_counts() <= 5000].index), 
                      'addr1'] = 'Others'
train_transaction.loc[train_transaction.\
                      addr2.isin(train_transaction.addr2.\
                                 value_counts()[train_transaction.addr2.\
                                                value_counts() <= 50].index), 
                      'addr2'] = 'Others'

def ploting_cnt_amt(DataFile, column, lim=2000):
    tmp = pd.crosstab(DataFile[column], DataFile['isFraud'], normalize='index') * 100
    tmp.reset_index(inplace=True)
    tmp.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)
    
    plt.figure(figsize=(16,15))
    plt.suptitle(f'{column}', fontsize=24)
    
    plt.subplot(211)
    plot_1 = sns.countplot(x=column, data=DataFile, order=list(tmp[column].values))
    plot_1_2 = plot_1.twinx()
    plot_1_2 = sns.pointplot(x=column, y='IsFraud', data=tmp, order=list(tmp[column].values),
                            color='black', legend=False)
    plot_1_2.set_ylim(0, tmp['IsFraud'].max()*1.1)
    plot_1_2.set_ylabel('异常消费占比', fontsize=14)
    
    plot_1.set_title(f'最常见的{column}的值与异常消费占比', fontsize=20)
    plot_1.set_xlabel(f'{column}的种类', fontsize=14)
    plot_1.set_ylabel('总计', fontsize=14)
    plot_1.set_xticklabels(plot_1.get_xticklabels(), rotation=45)
    
    
    sizes = []
    for p in plot_1.patches:
        height = p.get_height()
        sizes.append(height)
        plot_1.text(p.get_x() + p.get_width() / 2.,
                   height + 3,
                   '{:1.2f}%'.format(height / total * 100),
                   ha='center', fontsize=12)
    
    plot_1.set_ylim(0, max(sizes)*1.15)
    
    plt.subplot(212)
    perc_mat = (DataFile.groupby(['isFraud', column])['TransactionAmt'].sum()\
               /DataFile.groupby([column])['TransactionAmt'].sum()*100).unstack('isFraud')
    perc_mat.reset_index(inplace=True)
    perc_mat.rename(columns={0: 'NoFraud', 1: 'IsFraud'}, inplace=True)
    
    amt = DataFile.groupby([column])['TransactionAmt'].sum().reset_index()
    perc_mat = perc_mat.fillna(0)
    
    plot_2 = sns.barplot(x=column, y='TransactionAmt', 
                           data=amt, 
                           order=list(amt[column].values))
    plot_2_2 = plot_2.twinx()
    plot_2_2 = sns.pointplot(x=column, y='IsFraud', data=perc_mat,
                            order=list(amt[column].values),
                            color='black', legend=False)
    plot_2_2.set_ylim(0, perc_mat['IsFraud'].max()*1.1)
    plot_2_2.set_ylabel('异常消费占比', fontsize=14)
    
    plot_2.set_xticklabels(plot_2.get_xticklabels(), rotation=45)
    plot_2.set_title(f'{column} 各类消费金额 以及异常消费占比', fontsize=20)
    plot_2.set_xlabel(f'{column}种类', fontsize=14)
    plot_2.set_ylabel('消费金($)', fontsize=14)
    
    for p in plot_2.patches:
        height = p.get_height()
        plot_2.text(p.get_x() + p.get_width()/2.,
                   height + 3,
                   '{:1.2f}%'.format(height/total_amt*100),
                   ha='center', fontsize=12)
    
    
    plt.subplots_adjust(hspace=0.65, top=0.85, wspace=0.65)
    plt.show()

ploting_cnt_amt(train_transaction, 'addr1')

ploting_cnt_amt(train_transaction, 'addr2')

train_transaction.P_emaildomain.value_counts()[train_transaction.P_emaildomain.value_counts() <= 1000].index

CategoricalIndex(['charter.net', 'live.com.mx', 'rocketmail.com', 'mail.com',
                  'earthlink.net', 'gmail', 'outlook.es', 'mac.com',
                  'juno.com', 'aim.com', 'roadrunner.com', 'hotmail.es',
                  'windstream.net', 'hotmail.fr', 'frontier.com',
                  'embarqmail.com', 'web.de', 'twc.com', 'netzero.com',
                  'prodigy.net.mx', 'centurylink.net', 'netzero.net',
                  'frontiernet.net', 'q.com', 'suddenlink.net', 'cfl.rr.com',
                  'sc.rr.com', 'cableone.net', 'gmx.de', 'yahoo.fr',
                  'yahoo.es', 'hotmail.co.uk', 'protonmail.com', 'yahoo.de',
                  'ptd.net', 'live.fr', 'yahoo.co.uk', 'hotmail.de',
                  'servicios-ta.com', 'yahoo.co.jp'],
                 categories=['aim.com', 'anonymous.com', 'aol.com', 'att.net', 'bellsouth.net', 'cableone.net', 'centurylink.net', 'cfl.rr.com', ...], ordered=False, dtype='category')

train_transaction['P_emaildomain'] = train_transaction['P_emaildomain'].cat.add_categories(['NoInf','Others','Microsoft','Yahoo Mail','Google','Others'])

train_transaction.loc[train_transaction['P_emaildomain'].isin(['gmail.com', 'gmail']), 'P_emaildomain'] = 'Google'
train_transaction.loc[train_transaction['P_emaildomain'].isin(['yahoo.com', 'yahoo.com.mx', 'yahoo.co.uk', 
                                                              'yahoo.co.jp', 'yahoo.de','yahoo.fr','yahoo.es']), 
                      'P_emaildomain'] = 'Yahoo Mail'
train_transaction.loc[train_transaction['P_emaildomain'].isin(['hotmail.com', 'outlook.com', 'msn.com', 'live.com.mx',
                                                              'hotmail.es', 'hotmail.co.uk', 'hotmail.de',
                                                              'outlook.es', 'live.com', 'live.fr', 'hotmail.fr']),
                     'P_emaildomain'] = 'Microsoft'
train_transaction.loc[train_transaction.P_emaildomain.isin(train_transaction.\
                                                                    P_emaildomain.value_counts()[train_transaction.P_emaildomain.\
                                                                                                value_counts() <= 500]\
                                                                   .index), 'P_emaildomain'] = 'Others'
train_transaction.P_emaildomain.fillna('NoInf', inplace=True)

train_transaction['R_emaildomain'] = train_transaction['R_emaildomain'].cat.add_categories([0])

ploting_cnt_amt(train_transaction, 'R_emaildomain')

train_transaction.loc[train_transaction.C1.isin(train_transaction.C1.\
                                                value_counts()[train_transaction.C1.value_counts() <=400].index), 
                      'C1']= 'Others'

ploting_cnt_amt(train_transaction, 'C1')

train_transaction.loc[train_transaction.C2.isin(train_transaction.C2.\
                                                value_counts()[train_transaction.C2.value_counts() <=400].index), 
                      'C2']= 'Others'

ploting_cnt_amt(train_transaction, 'C2')

未完待续