加入VLM V 0.0.4, 2019 08 01

.gitignore

 tmp.py
 test.py
+.idea/
\ No newline at end of file

README.md

@@ -16,7 +16,7 @@
 
 # VLM
 
-- 待重写.
+- 重写完成, 代码流程改天写.
 
 # PSI & Lift Chart
 
@@ -49,9 +49,8 @@
   - 某个客群是否异常(AUC明显下降, PSI较大).
 - NOTE:
   - 当某月样本量很小, 或者没有样本时, 标记为NaN. 对应的PSI, AUC也为NaN.
-  - 当某月样本量比较小, 导致PSI, AUC计算异常(如某些分箱没有样本, 全为非逾期样本), 则标记为-999.
-  - 基准月的PSI为其与全样本第一个月之间的PSI, 反应该客群与整体的差别.
-  - 基准月的AUCR为其与全样本第一个月AUC之间的比率, 反应该客群与整体之间的差别.
+  - 当某月样本量比较小, 导致AUC计算异常(如某些分箱没有样本, 全为非逾期样本), 则标记为-999.
+  - 对于PSI的计算, 采用平滑处理, 根据PSI计算公式, 一旦某箱为0, 而基准月该分箱只要不为0, 就会导致计算异常, 因此计算PSI时为每箱的占比在原基础上增加0.001, 使计算更加稳定.
 
 ## 使用方法
 
@@ -60,7 +59,10 @@
 - 创建一个模型监控对象(这样我们就有对象了^v^).
 
   ```python
-  mm = ModelMonitor(excel_path='./model_score.xlsx', sheet_name='mongo_model', save_path='./image/', passdue_day=15, num_month=4, min_user_group=500, max_psi=0.1, min_aucr=0.8, if_load=False, if_save=True)
+  # 创建对象.
+  mm = ModelMonitor(excel_path='./model_score.xlsx', sheet_name='mongo_model', save_path='./image/', passdue_day=15, num_month=4, min_user_group=1000, max_psi=0.1, min_aucr=0.9, if_read=True, if_save=True)
+  # 执行run方法.
+  mm.run()
   ```
 
   - excel_path: Excel文件路径.
@@ -85,19 +87,14 @@
 
   - min_aucr: 最小AUCR, 小于则视为该客群异常.
 
-  - if_load: 是否从本地加载数据.
+  - if_read: 是否从数据库读取数据.
 
-    默认为False, 即从数据库抽取数据.
+    默认为False, 即从本地获取数据.
 
   - if_save: 是否将数据库抽取的数据保存到本地.
 
-- 执行run函数.
 
-  ```python
-  mm.run()
-  ```
-
-- 输出:
+## 输出
 
 - 图片保存在./image中.
   - PSI: ./image/PSI
@@ -109,9 +106,7 @@
 # TODO
 
 - 添加对存在MySQL中模型分计算PSI, AUC的代码.
-- 完成对VLM的重写.
-- 部分(量信分, app模型)模型分报错, 进一步与模型维护者交流, 看是否字段名或者其它地方有问题.
-- 对于某些客群在10个分箱中, 部分分箱数量为0导致计算异常的进行细致的处理.
+- 增加模型重要特征的VLM.
 
 # 版本信息
 
@@ -138,6 +133,13 @@
 
   - 增加数据读写模式的功能, 主要方便debug.
   
+- V 0.0.4
+
+  - 删除基准月AUCR, PSI的计算.
+  - 重写VLM.
+  - 增加过滤空跑数据的功能.
+  - 增强代码鲁棒性.
+
 # 贡献
 
 - 王家华 (jiahua.wang@quantgroup.cn)

model_monitor_PSI_AUC.py

@@ -23,8 +23,8 @@ warnings.filterwarnings('ignore')
 class ModelMonitor:
     def __init__(self, excel_path='./model_score.xlsx', sheet_name='mongo_model',
                  passdue_day=15, save_path='./image/',
-                 num_month=4, min_user_group=500, max_psi=0.1, min_aucr=0.85,
-                 if_save=True, if_load=False):
+                 num_month=4, min_user_group=1000, max_psi=0.1, min_aucr=0.85,
+                 if_save=True, if_read=True):
 
         # 考虑到数据库配置基本不变, 所以不设置创建对象时对应输入变量.
         self.mysql_engine = pymysql.connect(host='172.20.6.9',
@@ -40,10 +40,17 @@ class ModelMonitor:
         self.mongo_table = self.mongo_db['wf_audit_log_with_feature']
 
         # 读取整理在Excel中的模型相关信息.
-        self.model_info_df = pd.read_excel(excel_path, sheet_name=sheet_name)
-        self.model_name_list = self.model_info_df.model_name.tolist()
-        self.model_feild_list = self.model_info_df.model_feild.tolist()
-        self.model_feild_name_dict = dict(zip(self.model_feild_list, self.model_name_list))
+        self.field_info_df = pd.read_excel(excel_path, sheet_name=sheet_name)
+        self.field_name_list = self.field_info_df.field_name.tolist()
+        self.field_query_list = self.field_info_df.field_query.tolist()
+        self.field_app_type_list = self.field_info_df.app_type.tolist()
+        self.field_app_type_list = [str(x) for x in self.field_app_type_list]
+        self.field_query_name_dict = dict(zip(self.field_query_list, self.field_name_list))
+        self.field_query_app_type_dict = dict(zip(self.field_query_list, self.field_app_type_list))
+        ## 空跑信息.
+        self.na_time = self.field_info_df.na_time.tolist()  # 空跑时间段
+        self.na_app_type = self.field_info_df.na_app_type.tolist()  # 空跑申请类型
+        self.na_app_chan = self.field_info_df.na_app_chan.tolist()  # 空跑渠道
 
         # 一些定义的常量
         self.passdue_day = passdue_day  # 逾期天数, 默认15.
@@ -56,10 +63,11 @@ class ModelMonitor:
         # 获取当天日期信息.
         self.current_date = (datetime.date.today() + relativedelta(days=-1)).strftime('%Y-%m-%d')
         self.response_date = (datetime.date.today() + relativedelta(days=-(31 + self.passdue_day))).strftime('%Y-%m-%d')
-        self.first_date = (datetime.date.today() + relativedelta(months=-self.num_month + 1)).strftime('%Y-%m-01')
+        self.first_date = (datetime.date.today() + relativedelta(days=-1) + relativedelta(
+            months=-self.num_month + 1)).strftime('%Y-%m-01')
 
         self.current_month = (datetime.date.today() + datetime.timedelta(days=-1)).month
-        self.response_month = (datetime.date.today() + relativedelta(days=-46)).month
+        self.response_month = (datetime.date.today() + relativedelta(days=-(31 + self.passdue_day))).month
         self.first_month = self.current_month - self.num_month + 1
 
         # 将会从数据库中读取的数据.
@@ -83,12 +91,14 @@ class ModelMonitor:
         self.na_enough_data_psi_set = set()  # 一些新的模型没有足够数据用于统计.
         self.na_enough_data_auc_set = set()  # 一些新的模型没有足够数据用于统计.
 
-        self.filed_bench_bins_ratio = None  # 每个模型分在总体样本上第一个月的分箱比例.
-        self.filed_bench_auc = None  # 每个模型分在总体样本上第一个月的AUC
-
         # 程序数据读写模式.
         self.if_save = if_save  # 是否保存从数据库抽取的数据.
-        self.if_load = if_load  # 是否从保存的数据加载数据, 而不从数据库读取.
+        self.if_read = if_read  # 是否从从数据库读取.
+
+        # 分箱方式.
+        self.bins = None
+        self.bench_month = None
+
 
     def sql_query(self, sql):
         '''
@@ -136,8 +146,10 @@ class ModelMonitor:
             bins = score_list.quantile([.1, .2, .3, .4, .5, .6, .7, .8, .9]).values.tolist()
             bins = [-99999999] + bins + [99999999]
             bins = [x for x in bins if pd.notna(x)]
-            if len(set(bins)) < 11:
-                return None
+            bins = list(sorted(list(set(bins))))
+            # print(bins)
+            # if len(set(bins)) < 11:
+            #     return None
             return bins
         except:
             print('分箱出现错误.')
@@ -152,12 +164,68 @@ class ModelMonitor:
         :param array_2: array
         :return: PSI
         '''
+        # 对array做平滑处理, 防止一些分箱为零的PSI计算异常.
+        array_1 = array_1 + 0.001
+        array_2 = array_2 + 0.001
         try:
             psi = ((array_1 - array_2) * np.log10(array_1 / array_2)).sum()
             return psi
         except:
             return None
 
+    def filter_data(self, df, field):
+        '''
+        过滤空跑数据.
+        :param df: df.
+        :param field: str, 字段名.
+        :return: df, 过滤后的数据.
+        '''
+        df = df[~((df['applied_type'] == 1) & (df['applied_channel'].apply(lambda x: 'Android' in x)))]
+
+        field_idx = self.field_query_list.index(field)
+        na_time = self.na_time[field_idx]
+        na_type = self.na_app_type[field_idx]
+        na_chan = self.na_app_chan[field_idx]
+        print(na_chan, type(na_chan))
+        if pd.isnull(na_time):  # 没有空跑时间, 则不记录.
+            return df
+        # 时间.
+        t_s, t_e = na_time.split('~')
+        print(t_s, t_e)
+        if len(t_e) == 0:  # 若还在空跑, 则不记录.
+            return pd.DataFrame()
+        else:
+            print(df['applied_at'].head())
+            na_df = df[
+                (df['applied_at'].apply(lambda x: x[:10] >= t_s)) & (df['applied_at'].apply(lambda x: x[:10] <= t_e))]
+            if na_df.shape[0] == 0:
+                return df
+        # 申请类型.
+        if pd.isnull(na_type):
+            return df[~df.index.isin(na_df.index.values)]
+        else:
+            tmp_df = pd.DataFrame()
+
+            for i in str(int(na_type)):
+                print(i, 'wsnd')
+                tmp_df = tmp_df.append(na_df[na_df['applied_type'] == int(i)])
+            na_df = tmp_df
+            if na_df.shape[0] == 0:
+                return df
+        # 申请渠道.
+        if pd.isnull(na_chan):
+            return df[~df.index.isin(na_df.index.values)]
+        else:
+            tmp_df = pd.DataFrame()
+            for i in na_chan.split(','):
+                tmp_df = tmp_df.append(na_df[na_df['applied_channel'].apply(lambda x: i in x)])
+            na_df = tmp_df
+            if na_df.shape[0] == 0:
+                return df
+
+        print(df.shape[0], na_df.shape[0])
+        return df[~df.index.isin(na_df.index.values)]
+
     def helper_psi(self, user_group_name=None, df=None, info_dict=None, field=None):
         '''
         信息提取函数.
@@ -172,35 +240,31 @@ class ModelMonitor:
             month_list.remove(0)
 
         df_g = df.groupby(['month_label', 'bins']).agg({field: ['count']})
-
+        df_g.columns = ['_'.join(x) for x in df_g.columns.ravel()]
         df_g = df_g.reset_index()
         df_g = df_g.sort_values(['month_label', 'bins'])
         for i, m in enumerate(month_list):
-            amt_in_bins = df_g.loc[df_g['month_label'] == m, field].values
+            if m < self.bench_month:
+                continue
+            amt_in_bins = df_g.loc[df_g['month_label'] == m, ['bins', field + '_count']]
+            amt_in_bins = pd.merge(left=self.bins, right=amt_in_bins, on='bins', how='left')
+            amt_in_bins[field + '_count'] = amt_in_bins[field + '_count'].fillna(0)
+            amt_in_bins = amt_in_bins[field + '_count'].values
+
             ## 某月样本量小于阈值, 放弃记录信息.
             if amt_in_bins.sum() < self.min_user_group:
                 print('%d月样本量过小, 放弃提取信息.' % m)
                 continue
+
             info_dict[user_group_name][str(m) + '月'] = {}
             info_dict[user_group_name][str(m) + '月']['该月样本量'] = amt_in_bins.sum()
             info_dict[user_group_name][str(m) + '月']['各分箱样本量'] = amt_in_bins
-            info_dict[user_group_name][str(m) + '月']['各分箱样本占比'] = np.array([x[0] for x in amt_in_bins / amt_in_bins.sum()])
-            ## 若为某模型分全样本, 则记录分箱比例.
-            if user_group_name == '全样本' and self.filed_bench_bins_ratio is None:
-                self.filed_bench_bins_ratio = info_dict[user_group_name][str(m) + '月']['各分箱样本占比']
+            info_dict[user_group_name][str(m) + '月']['各分箱样本占比'] = amt_in_bins / amt_in_bins.sum()
             print('%d月样本量: %d' % (m, info_dict[user_group_name][str(m) + '月']['该月样本量']))
         # 计算PSI, 以样本量达标的第一个月为基准.
         for i, m in enumerate(info_dict[user_group_name]):
             if i == 0:
-                if user_group_name == '全样本':
-                    psi = 0
-                else:
-                    psi = self.calc_psi(self.filed_bench_bins_ratio, info_dict[user_group_name][m]['各分箱样本占比'])
-                if psi is not None:
-                    info_dict[user_group_name][m]['psi'] = psi
-                else:
-                    info_dict[user_group_name][m]['psi'] = -999
-                    print('计算PSI出现错误.')
+                info_dict[user_group_name][m]['psi'] = 0
                 bench_bins_ratio = info_dict[user_group_name][m]['各分箱样本占比']
             else:
                 psi = self.calc_psi(bench_bins_ratio, info_dict[user_group_name][m]['各分箱样本占比'])
@@ -231,19 +295,27 @@ class ModelMonitor:
         df_g = df_g.sort_values(['month_label', 'bins'])
 
         for i, m in enumerate(month_list):
-            amt_in_bins = df_g.loc[df_g['month_label'] == m, 'overdue_count'].values
+            if m < self.bench_month:
+                continue
+            amt_in_bins = df_g.loc[df_g['month_label'] == m, ['bins', 'overdue_count']]
+            amt_in_bins = pd.merge(left=self.bins, right=amt_in_bins, on='bins', how='left')
+            amt_in_bins['overdue_count'] = amt_in_bins['overdue_count'].fillna(0)
+            amt_in_bins = amt_in_bins['overdue_count'].values
+
             # 某月样本量小于阈值, 放弃记录信息.
             if amt_in_bins.sum() < self.min_user_group:
                 print('%d月样本量过小, 放弃提取信息.' % m)
                 continue
             info_dict[user_group_name][str(m) + '月'] = {}
-
             info_dict[user_group_name][str(m) + '月']['该月样本量'] = amt_in_bins.sum()
             info_dict[user_group_name][str(m) + '月']['各分箱样本量'] = amt_in_bins
-            info_dict[user_group_name][str(m) + '月']['各分箱逾期样本量'] = df_g.loc[
-                df_g['month_label'] == m, 'overdue_sum'].values
-            info_dict[user_group_name][str(m) + '月']['各分箱逾期率'] = df_g.loc[
-                df_g['month_label'] == m, 'overdue_mean'].values
+            overdue_in_bins = df_g.loc[df_g['month_label'] == m, ['bins', 'overdue_sum']]
+            overdue_in_bins = pd.merge(left=self.bins, right=overdue_in_bins, on='bins', how='left')
+            overdue_in_bins['overdue_sum'] = overdue_in_bins['overdue_sum'].fillna(0)
+            overdue_in_bins = overdue_in_bins['overdue_sum'].values
+            info_dict[user_group_name][str(m) + '月']['各分箱逾期样本量'] = overdue_in_bins
+            info_dict[user_group_name][str(m) + '月']['各分箱逾期率'] = np.array(
+                [x if ~np.isnan(x) else 0 for x in overdue_in_bins / amt_in_bins])
             print('%d月样本量: %d' % (m, info_dict[user_group_name][str(m) + '月']['该月样本量']))
             try:
                 info_dict[user_group_name][str(m) + '月']['auc'] = roc_auc_score(
@@ -252,12 +324,9 @@ class ModelMonitor:
             except:
                 print('AUC计算发生错误.')
                 info_dict[user_group_name][str(m) + '月']['auc'] = -999
-            if user_group_name == '全样本' and self.filed_bench_auc is None:
-                self.filed_bench_auc = info_dict[user_group_name][str(m) + '月']['auc']
-
         for i, m in enumerate(info_dict[user_group_name]):
             if i == 0:  # 基准月.
-                info_dict[user_group_name][m]['aucR'] = info_dict[user_group_name][m]['auc'] / self.filed_bench_auc
+                info_dict[user_group_name][m]['aucR'] = 1
                 bench_month = m
             else:
                 info_dict[user_group_name][m]['aucR'] = info_dict[user_group_name][m]['auc'] / \
@@ -273,22 +342,40 @@ class ModelMonitor:
         if not os.path.exists(self.save_path + 'PSI/'):
             os.mkdir(self.save_path + 'PSI/')
         # 分离数据.
-        df_copy = self.merge_data[[field, 'month_label', 'applied_type', 'applied_channel']].copy()
-
+        df_copy = self.merge_data[[field, 'month_label', 'applied_type', 'applied_channel', 'applied_at']].copy()
+
+        # 选择包含正确申请类型的数据.
+        tmp_df = pd.DataFrame()
+        for i in self.field_query_app_type_dict[field]:
+            tmp_df = tmp_df.append(df_copy[df_copy['applied_type'] == int(i)])
+        df_copy = tmp_df
+        df_copy = df_copy[df_copy[field].notna()]
+        # 过滤空跑数据.
+        df_copy = self.filter_data(df_copy, field)
+        if df_copy.shape[0] == 0:
+            print('仍在空跑.')
+            return None
+        print(df_copy['applied_at'].min(), df_copy['applied_at'].max())
         # 对模型分进行分箱, 选取数据中该模型分最开始的那个月作为基准.
+        bins = None
         for m in range(self.first_month, self.current_month + 1):
+            if df_copy.loc[df_copy['month_label'] == m, field].shape[0] < self.min_user_group:
+                continue
+            else:
                 bins = self.make_bin(df_copy.loc[df_copy['month_label'] == m, field])
                 if bins:
-                print('%s以%d月为基准月.' % (self.model_feild_name_dict[field], m))
+                    print('%s以%d月为基准月.' % (self.field_query_name_dict[field], m))
+                    self.bench_month = m
                     break
         if not bins:
-            self.na_enough_data_psi_set.add(self.model_feild_name_dict[field])
-            print('%s 数据时间跨度不足, 放弃画图.' % self.model_feild_name_dict[field])
+            self.na_enough_data_psi_set.add(self.field_query_name_dict[field])
+            print('%s 数据时间跨度不足, 放弃画图.' % self.field_query_name_dict[field])
             print('=' * 40)
             return None
 
-        df_copy['bins'] = pd.cut(df_copy[field], bins)  # 根据分箱规则进行分箱.
-
+        df_copy['bins'] = pd.cut(df_copy[field], bins, precision=8)  # 根据分箱规则进行分箱.
+        self.bins = pd.Series(df_copy['bins'].unique(), name='bins').sort_values()
+        self.bins = self.bins.dropna()
         # 包含各种信息的字典.
         # 如: {'全样本':
         #               {'3月':
@@ -317,15 +404,17 @@ class ModelMonitor:
         app_type_dict = {1: '首申', 2: '复申', 3: '复贷'}
         df_copy_g = df_copy.groupby(['applied_type', 'applied_channel'])[field].count().sort_values(ascending=False)
         df_copy_g = df_copy_g.reset_index()
+
         ## 过滤小客群.
         df_copy_g = df_copy_g.loc[df_copy_g[field] > self.min_user_group * self.num_month]
-        app_type_set = df_copy_g['applied_type'].unique()
-        app_chan_set = df_copy_g['applied_channel'].unique()
-        for app_type in app_type_set:
-            for app_chan in app_chan_set:
-                if df_copy_g.loc[
-                    (df_copy_g['applied_type'] == app_type) & (df_copy_g['applied_channel'] == app_chan)].shape[0] != 0:
-                    user_group_dict[app_type_dict[app_type] + '-' + app_chan] = (app_type, app_chan)
+
+        tmp_df = pd.DataFrame()
+        for i in range(df_copy_g.shape[0]):
+            tmp_df = tmp_df.append(df_copy.loc[(df_copy['applied_type'] == df_copy_g.iloc[i]['applied_type']) &
+                                               (df_copy['applied_channel'] == df_copy_g.iloc[i]['applied_channel'])])
+            user_group_dict[
+                app_type_dict[df_copy_g.iloc[i]['applied_type']] + '-' + df_copy_g.iloc[i]['applied_channel']] = \
+                (df_copy_g.iloc[i]['applied_type'], df_copy_g.iloc[i]['applied_channel'])
         del df_copy_g
         ## 按划分的客群处理数据.
         for user_group_name in user_group_dict:
@@ -346,7 +435,7 @@ class ModelMonitor:
         print('开始画图.')
         print('=' * 40)
         for user_group_name in info_dict:
-            print(self.model_feild_name_dict[field] + '-' + user_group_name)
+            print(self.field_query_name_dict[field] + '-' + user_group_name)
             plt.figure(figsize=(16, 8))
             for m in info_dict[user_group_name]:
                 # print(m)
@@ -357,14 +446,14 @@ class ModelMonitor:
                          label='%s PSI: %.3f \n 样本量: %d' % (
                              m, info_dict[user_group_name][m]['psi'], info_dict[user_group_name][m]['该月样本量']))
             plt.legend(loc='upper right')
-            plt.title(self.model_feild_name_dict[field] + '-' + user_group_name)
-            plt.savefig(self.save_path + 'PSI/' + self.model_feild_name_dict[field] + '-' + user_group_name)
+            plt.title(self.field_query_name_dict[field] + '-' + user_group_name)
+            plt.savefig(self.save_path + 'PSI/' + self.field_query_name_dict[field] + '-' + user_group_name)
             plt.show()
 
         # 保存统计信息.
         for user_group_name in info_dict:
             # print(self.model_feild_name_dict[field] + '-' + user_group_name)
-            tmp_dict = {'模型名称': [self.model_feild_name_dict[field]],
+            tmp_dict = {'模型名称': [self.field_query_name_dict[field]],
                         '客群名称': [user_group_name]}
             for m in info_dict[user_group_name]:
                 tmp_dict[m[0] + '月数量'] = [int(info_dict[user_group_name][m]['该月样本量'])]
@@ -380,6 +469,16 @@ class ModelMonitor:
         # 分离数据.
         df_copy = self.merge_data[
             [field, 'month_label', 'applied_type', 'applied_channel', 'overdue', 'passdue_day', 'applied_at']].copy()
+        df_copy = df_copy[(df_copy['overdue'].notna()) & (df_copy[field].notna())]
+        tmp_df = pd.DataFrame()
+        for i in self.field_query_app_type_dict[field]:
+            tmp_df = tmp_df.append(df_copy[df_copy['applied_type'] == int(i)])
+        df_copy = tmp_df
+        # 过滤空跑数据.
+        df_copy = self.filter_data(df_copy, field)
+        if df_copy.shape[0] == 0:
+            print('仍在空跑.')
+            return None
         ## 筛选出放款, 且逾期表现的数据.
         if repr(df_copy['applied_at'].dtype) == "dtype('O')":
             df_copy = df_copy.loc[
@@ -391,18 +490,25 @@ class ModelMonitor:
                                           df_copy[field] > 0) & (df_copy['passdue_day'].notna())]
 
         # 对模型分进行分箱, 选取数据中该模型分最开始的那个月作为基准.
+        bins = None
         for m in range(self.first_month, self.response_month + 1):
+            if df_copy.loc[df_copy['month_label'] == m, field].shape[0] < self.min_user_group:
+                continue
+            else:
                 bins = self.make_bin(df_copy.loc[df_copy['month_label'] == m, field])
                 if bins:
-                print('%s以%d月为基准月.' % (self.model_feild_name_dict[field], m))
+                    print('%s以%d月为基准月.' % (self.field_query_name_dict[field], m))
+                    self.bench_month = m
                     break
         if not bins:
-            self.na_enough_data_auc_set.add(self.model_feild_name_dict[field])
-            print('%s 数据时间跨度不足, 放弃画图.' % self.model_feild_name_dict[field])
+            self.na_enough_data_auc_set.add(self.field_query_name_dict[field])
+            print('%s 数据时间跨度不足, 放弃画图.' % self.field_query_name_dict[field])
             print('=' * 40)
             return None
 
-        df_copy['bins'] = pd.cut(df_copy[field], bins)  # 根据分箱规则进行分箱.
+        df_copy['bins'] = pd.cut(df_copy[field], bins, precision=8)  # 根据分箱规则进行分箱.
+        self.bins = pd.Series(df_copy['bins'].unique(), name='bins').sort_values()
+        self.bins = self.bins.dropna()
 
         # 包含各种信息的字典.
         # 如: {'全样本':
@@ -465,7 +571,7 @@ class ModelMonitor:
         print('开始画图.')
         print('=' * 40)
         for user_group_name in info_dict:
-            print(self.model_feild_name_dict[field] + '-' + user_group_name)
+            print(self.field_query_name_dict[field] + '-' + user_group_name)
             plt.figure(figsize=(16, 8))
             for m in info_dict[user_group_name]:
                 ## 若某月数量少, 则放弃画图.
@@ -478,13 +584,13 @@ class ModelMonitor:
                              m, info_dict[user_group_name][m]['auc'], info_dict[user_group_name][m]['aucR'],
                              info_dict[user_group_name][m]['该月样本量']))
             plt.legend(loc='upper right')
-            plt.title(self.model_feild_name_dict[field] + '-' + user_group_name)
-            plt.savefig(self.save_path + 'AUC/' + self.model_feild_name_dict[field] + '-' + user_group_name)
+            plt.title(self.field_query_name_dict[field] + '-' + user_group_name)
+            plt.savefig(self.save_path + 'AUC/' + self.field_query_name_dict[field] + '-' + user_group_name)
             plt.show()
 
         # 保存统计信息.
         for user_group_name in info_dict:
-            tmp_dict = {'模型名称': [self.model_feild_name_dict[field]],
+            tmp_dict = {'模型名称': [self.field_query_name_dict[field]],
                         '客群名称': [user_group_name]}
             for m in info_dict[user_group_name]:
                 tmp_dict[m[0] + '月数量'] = [int(info_dict[user_group_name][m]['该月样本量'])]
@@ -495,16 +601,8 @@ class ModelMonitor:
 
     def abnormal_psi(self):
         def is_abnormal_psi(data):
-            first = True
             for idx in data.index:
-                if 'PSI' in idx and pd.notna(data[idx]):
-                    if first:
-                        if data[idx] > self.max_psi * 5:
-                            return True
-                        else:
-                            first = False
-                    else:
-                        if data[idx] > self.max_psi:
+                if 'PSI' in idx and pd.notna(data[idx]) and data[idx] > self.max_psi:
                     return True
             return False
 
@@ -513,7 +611,7 @@ class ModelMonitor:
     def abnormal_auc(self):
         def is_abnormal_auc(data):
             for i in data.index:
-                if 'AUCR' in i and data[i] < self.min_aucr:
+                if 'AUCR' in i and pd.notna(data[i]) and data[i] < self.min_aucr:
                     return True
             return False
 
@@ -521,32 +619,41 @@ class ModelMonitor:
 
     def run(self):
         # 获取MySQL数据, 取近期num_month个月数据(如今天7.27, 则这27天算进7月).
-        if not self.if_load:
+        if self.if_read:
             self.mysql_df = self.sql_query('''SELECT order_no, applied_at,
-                                            applied_type, applied_from, applied_channel, transacted, passdue_day
+                                            applied_from, applied_channel, transacted, passdue_day
                                             FROM risk_analysis
                                             WHERE applied_at >= "%s 00:00:00"
                                             AND applied_at <= "%s 00:00:00"'''
-                                           % (self.first_date, self.current_date))
+                                           % (self.first_date, datetime.date.today().strftime('%Y-%m-%d')))
             print('MySQL数据获取成功.')
             if self.if_save:
                 self.mysql_df.to_csv('./mysql_data.csv', index=False)
         else:
             self.mysql_df = pd.read_csv('./mysql_data.csv')
 
+        def func_0(data):
+            try:
+                return int(int(data) + 1)
+            except:
+                return np.nan
+
         # 获取MongoDB数据, 取近期num_month个月数据(如今天7.27, 则这27天算进7月).
-        if not self.if_load:
+        if self.if_read:
             condition = {'wf_created_at': {'$gte': '%s 00:00:00' % self.first_date,
-                                           '$lte': '%s 00:00:00' % self.current_date}}
-            fields = {'wf_biz_no': 1, 'wf_created_at': 1}
-            for f in self.model_feild_list:  # 加入Excel中预置的模型分名称
+                                           '$lte': '%s 00:00:00' % datetime.date.today().strftime('%Y-%m-%d')}}
+            fields = {'wf_biz_no': 1, 'wf_created_at': 1, 'wf_loan_type': 1}
+            for f in self.field_query_list:  # 加入Excel中预置的模型分名称
                 fields[f] = 1
             self.mongo_df = self.mongo_query(condition, fields)
+            self.mongo_df['applied_type'] = self.mongo_df['wf_loan_type'].apply(func_0)
+            del self.mongo_df['wf_loan_type']
             print('MongoDB数据获取成功.')
             if self.if_save:
                 self.mongo_df.to_csv('./mongo_data.csv', index=False)
         else:
             self.mongo_df = pd.read_csv('./mongo_data.csv')
+        self.mongo_df = self.mongo_df.loc[self.mongo_df['applied_type'].notna()]
 
         # MySQL数据去重.
         self.mysql_df = self.mysql_df.sort_values('passdue_day')
@@ -555,7 +662,8 @@ class ModelMonitor:
 
         # 拼接数据.
         self.merge_data = pd.merge(left=self.mysql_df, right=self.mongo_df,
-                                   left_on='order_no', right_on='wf_biz_no', how='left')
+                                   left_on='order_no', right_on='wf_biz_no', how='inner')
+        print('数据拼接完成.')
 
         ## 定义逾期用户.
         def overdue(data):
@@ -566,6 +674,13 @@ class ModelMonitor:
 
         self.merge_data['overdue'] = self.merge_data['passdue_day'].apply(overdue)
 
+
+        # 清洗时间格式, 使其转换成统一的字符串格式.
+        if repr(self.merge_data['applied_at'].dtype) == "dtype('O')":
+            self.merge_data['applied_at'] = self.merge_data['applied_at'].apply(lambda x: x[:10])
+        else:
+            self.merge_data['applied_at'] = self.merge_data['applied_at'].apply(lambda x: x.strftime('%Y-%m-%d'))
+
         # 清洗数据.
         def clean_data(data):
             try:
@@ -574,20 +689,19 @@ class ModelMonitor:
                 return np.nan
 
         na_field_list = []
-        for field in self.model_feild_list:
+        for field in self.field_query_list:
             if field in self.merge_data.columns.tolist():
-                print('正在清洗%s' % self.model_feild_name_dict[field])
+                print('正在清洗%s' % self.field_query_name_dict[field])
                 self.merge_data[field] = self.merge_data[field].apply(clean_data)
             else:
                 na_field_list.append(field)
         ## 去除因为一些原因未抽取到的字段.
         print('不包含以下字段:')
         for field in na_field_list:
-            print(self.model_feild_name_dict[field])
-            self.model_feild_list.remove(field)
-            self.model_name_list.remove(self.model_feild_name_dict[field])
-            del self.model_feild_name_dict[field]
-        print('数据拼接完成.')
+            print(self.field_query_name_dict[field])
+            self.field_query_list.remove(field)
+            self.field_name_list.remove(self.field_query_name_dict[field])
+            del self.field_query_name_dict[field]
 
         # 数据按月划分.
         self.merge_data['month_label'] = 0
@@ -601,11 +715,24 @@ class ModelMonitor:
 
         ## PSI.
         print('开始画图-PSI.')
-        for field in self.model_feild_list:
+        for field in self.field_query_list:
+            # if self.field_query_name_dict[field] == '新颜v3':
+            #     self.plot_psi(field)
+            # if self.field_query_name_dict[field] == '短信分':
+            #     self.plot_psi(field)
+            # if self.field_query_name_dict[field] == '探知':
+            #     self.plot_psi(field)
+
             self.plot_psi(field)
+
         ## AUC.
         print('开始画图-AUC.')
-        for field in self.model_feild_list:
+        for field in self.field_query_list:
+        #     if self.model_feild_name_dict[field] == '短信分':
+        #         self.plot_auc(field)
+        #     if self.model_feild_name_dict[field] == '百融v1':
+        #         self.plot_auc(field)
+
             self.plot_auc(field)
 
         # 输出数据不足的模型.

model_monitor_VLM.py

+# coding=utf-8
+import pandas as pd
+import numpy as np
+
+import matplotlib.pyplot as plt
+
+import os
+import pickle
+import datetime
+from dateutil.relativedelta import relativedelta
+from collections import OrderedDict
+from scipy.stats import norm
+
+import pymysql
+import pymongo
+
+import warnings
+
+warnings.filterwarnings('ignore')
+plt.rcParams['font.sans-serif'] = ['SimHei']
+plt.rcParams['axes.unicode_minus'] = False
+plt.rcParams['savefig.dpi'] = 100
+
+
+class ModelMonitorVLM:
+    def __init__(self, excel_path='./model_score.xlsx',
+                 sheet_name='mongo_model',
+                 fig_save_path='./image/', info_save_path='./info/',
+                 data_save_path='./data/',
+                 if_read=True, if_save=True,
+                 alpha=0.01, min_user_group=10000):
+
+        # 考虑到数据库配置基本不变, 所以不设置创建对象时对应输入变量.
+        self.mysql_engine = pymysql.connect(host='172.20.6.9',
+                                            port=9030,
+                                            user='fengkong_read_only',
+                                            passwd='mT2HFUgI',
+                                            db='risk_analysis',
+                                            charset='utf8')
+
+        self.mongo_client = pymongo.MongoClient(
+            "mongodb://haoyue.shu:x2egwRHk7WhQ4So1@172.18.3.22:27017/?authSource=rc_mgo_feature_dp")
+        self.mongo_db = self.mongo_client['rc_mgo_feature_dp']
+        self.mongo_table = self.mongo_db['wf_audit_log_with_feature']
+
+        # 读取整理在Excel中的模型相关信息.
+        self.field_info_df = pd.read_excel(excel_path, sheet_name=sheet_name)
+        self.field_name_list = self.field_info_df.field_name.tolist()
+        self.field_query_list = self.field_info_df.field_query.tolist()
+        self.field_app_type_list = self.field_info_df.app_type.tolist()
+        self.field_app_type_list = [str(x) for x in self.field_app_type_list]
+        self.field_query_name_dict = dict(zip(self.field_query_list, self.field_name_list))
+        self.field_query_app_type_dict = dict(zip(self.field_query_list, self.field_app_type_list))
+        ## 空跑信息.
+        self.na_time = self.field_info_df.na_time.tolist()  # 空跑时间段
+        self.na_app_type = self.field_info_df.na_app_type.tolist()  # 空跑申请类型
+        self.na_app_chan = self.field_info_df.na_app_chan.tolist()  # 空跑渠道
+
+        # 文件存储路径.
+        self.fig_save_path = fig_save_path
+        self.info_save_path = info_save_path
+        self.data_save_path = data_save_path
+        if not os.path.exists(fig_save_path):
+            os.mkdir(fig_save_path)
+        if not os.path.exists(fig_save_path + 'image/'):
+            os.mkdir(fig_save_path + 'image/')
+        if not os.path.exists(fig_save_path + 'over_3std/'):
+            os.mkdir(fig_save_path + 'over_3std/')
+        if not os.path.exists(fig_save_path + 'trend/'):
+            os.mkdir(fig_save_path + 'trend/')
+        if not os.path.exists(info_save_path):
+            os.mkdir(info_save_path)
+        if not os.path.exists(data_save_path):
+            os.mkdir(data_save_path)
+
+        # MK test显著性水平.
+        self.alpha = alpha
+
+        # 数据.
+        self.mysql_df = None
+        self.mongo_df = None
+        self.merge_data = None
+
+        # 数据模式.
+        self.if_read = if_read
+        self.if_save = if_save
+
+        # 获取当天日期信息.
+        self.current_date = (datetime.date.today() + relativedelta(days=-1)).strftime('%Y-%m-%d')
+        self.third_date = datetime.date.today().strftime('%Y-%m-01')
+        self.second_date = (datetime.date.today() + relativedelta(months=-1)).strftime('%Y-%m-01')
+        self.first_date = (datetime.date.today() + relativedelta(months=-2)).strftime('%Y-%m-01')
+
+        self.current_month = (datetime.date.today() + datetime.timedelta(days=-1)).month
+        self.second_month = (datetime.date.today() + relativedelta(months=-1)).month
+        self.first_month = (datetime.date.today() + relativedelta(months=-2)).month
+
+        self.num_day = len(pd.date_range(self.first_date, self.current_date))
+
+        # 统计信息.
+        # {group_name_0: {time_list: [...],
+        #                 value_list: [...],
+        #                 count_list: [...],
+        #                 miss_rate_list: [...],
+        #                 zero_rate_list: [...],
+        #                 mean: float,
+        #                 std: float,
+        #                 trend: [...],
+        #                 h: [...],
+        #                 p: [...],
+        #                 over_3std: bool}
+        self.vlm_info_dict = OrderedDict()
+        self.vlm_info_df = None
+
+        self.min_user_group = min_user_group
+
+    def query_mysql(self, sql):
+        '''
+        连接MySQL数据库, 根据SQL返回数据.
+        :param sql: str.
+        :return: DataFrame.
+        '''
+        try:
+            return pd.read_sql(sql, self.mysql_engine)
+        except:
+            print('SQL查询出现错误.')
+        pass
+
+    def query_mongo(self, condition, fields):
+        '''
+        连接MongoDB, 根据查询返回数据.
+        :param condition: dict
+        :param fields: dict
+        :return: DataFrame
+        '''
+        try:
+            return pd.DataFrame(list(self.mongo_table.find(condition, fields)))
+        except:
+            print('Mongo查询出现错误.')
+
+        pass
+
+    def filter_data(self, df, field):
+        '''
+        过滤空跑数据.
+        :param df: df.
+        :param field: str, 字段名.
+        :return: df, 过滤后的数据.
+        '''
+
+        df = df[~((df['applied_type'] == 1) & (df['applied_channel'].apply(lambda x: 'Android' in x)))]
+        field_idx = self.field_query_list.index(field)
+        na_time = self.na_time[field_idx]
+        na_type = self.na_app_type[field_idx]
+        na_chan = self.na_app_chan[field_idx]
+        print(na_chan, type(na_chan))
+        if pd.isnull(na_time):  # 没有空跑时间, 则不记录.
+            return df
+        # 时间.
+        t_s, t_e = na_time.split('~')
+        print(t_s, t_e)
+        if len(t_e) == 0:  # 若还在空跑, 则不记录.
+            return pd.DataFrame()
+        else:
+            print(df['applied_at'].head())
+            na_df = df[
+                (df['applied_at'].apply(lambda x: x[:10] >= t_s)) & (df['applied_at'].apply(lambda x: x[:10] <= t_e))]
+            if na_df.shape[0] == 0:
+                return df
+        # 申请类型.
+        if pd.isnull(na_type):
+            return df[~df.index.isin(na_df.index.values)]
+        else:
+            tmp_df = pd.DataFrame()
+
+            for i in str(int(na_type)):
+                print(i, 'wsnd')
+                tmp_df = tmp_df.append(na_df[na_df['applied_type'] == int(i)])
+            na_df = tmp_df
+            if na_df.shape[0] == 0:
+                return df
+        # 申请渠道.
+        if pd.isnull(na_chan):
+            return df[~df.index.isin(na_df.index.values)]
+        else:
+            tmp_df = pd.DataFrame()
+            for i in na_chan.split(','):
+                tmp_df = tmp_df.append(na_df[na_df['applied_channel'].apply(lambda x: i in x)])
+            na_df = tmp_df
+            if na_df.shape[0] == 0:
+                return df
+
+        print(df.shape[0], na_df.shape[0])
+        return df[~df.index.isin(na_df.index.values)]
+
+
+    def mk_test(self, x, alpha=0.01):
+        '''
+        MK test.
+        用于检测时间序列趋势.
+        :param x: list[float]
+        :param alpha: float, 显著性水平.
+        :return: 趋势, 是否显著, P值, Z统计值.
+        '''
+        if type(x) != list:
+            x = x.values.tolist()
+        n = len(x)
+
+        # calculate S
+        s = 0
+        for k in range(n - 1):
+            for j in range(k + 1, n):
+                s += np.sign(x[j] - x[k])
+
+        # calculate the unique data
+        unique_x = np.unique(x)
+        g = len(unique_x)
+
+        # calculate the var(s)
+        if n == g:  # there is no tie
+            var_s = (n * (n - 1) * (2 * n + 5)) / 18
+        else:  # there are some ties in data
+            tp = np.zeros(unique_x.shape)
+            for i in range(len(unique_x)):
+                tp[i] = sum(x == unique_x[i])
+            var_s = (n * (n - 1) * (2 * n + 5) - np.sum(tp * (tp - 1) * (2 * tp + 5))) / 18
+
+        if s > 0:
+            z = (s - 1) / np.sqrt(var_s)
+        elif s < 0:
+            z = (s + 1) / np.sqrt(var_s)
+        else:  # s == 0:
+            z = 0
+
+        # calculate the p_value
+        p = 2 * (1 - norm.cdf(abs(z)))  # two tail test
+        h = abs(z) > norm.ppf(1 - alpha / 2)
+
+        if (z < 0) and h:
+            trend = 'decreasing'
+        elif (z > 0) and h:
+            trend = 'increasing'
+        else:
+            trend = 'no trend'
+
+        return trend, h, p, z
+
+    def process_data_helper(self, group_name=None, df=None, field=None):
+        print('正在处理%s' % group_name)
+        # 得到一个连续时间序列.
+        date_list = pd.date_range(self.first_date, self.current_date).strftime('%Y-%m-%d').values.tolist()
+
+        # 将特殊值标定为缺失.
+        def set_na(data):
+            if pd.isnull(data):
+                return np.nan
+            if data < 0 or data > 999999:
+                return np.nan
+            return data
+        print(df.head())
+        df = df[['applied_at', field]]
+        df[field] = df[field].apply(set_na)
+
+        # 计算该字段在每天的均值, 数量, 缺失率, 零率.
+        def count(data):
+            return len(data)
+
+        def miss_rate(data):
+            return data.isnull().mean()
+
+        def zero_rate(data):
+            return (data == 0).mean()
+
+        df_g = df.groupby('applied_at').agg({field: ['mean', count, miss_rate, zero_rate]})
+        df_g.columns = ['_'.join(x) for x in df_g.columns.ravel()]
+        df_g = df_g.reset_index()
+
+        # 将没有数据的当天均值补为0.
+        tmp_df = pd.DataFrame({'applied_at': date_list})
+        df_g = pd.merge(left=tmp_df, right=df_g, on='applied_at', how='left')
+        df_g = df_g.fillna(0)
+
+        # 计算该字段整体的均值.
+        mean = df[field].mean()
+
+        # 计算该字段整体的标准差.
+        std = df_g[field + '_mean'].std()
+
+        # 计算该字段的趋势信息.
+        trend_0, h_0, p_0, z_0 = self.mk_test(df_g[field + '_mean'], alpha=self.alpha)  # 整体.
+        trend_1, h_1, p_1, z_1 = self.mk_test(df_g.loc[df_g['applied_at'] < self.second_date, field + '_mean'],
+                                              alpha=self.alpha)  # -2月份
+        trend_2, h_2, p_2, z_2 = self.mk_test(df_g.loc[(df_g['applied_at'] >= self.second_date) & (
+                df_g['applied_at'] < self.third_date), field + '_mean'], alpha=self.alpha)  # -1月份
+        if datetime.date.today().day > 15:  # 0月份
+            trend_3, h_3, p_3, z_3 = self.mk_test(df_g.loc[df_g['applied_at'] >= self.third_date, field + '_mean'],
+                                                  alpha=self.alpha)
+        else:
+            trend_3, h_3, p_3, z_3 = 'NaN', np.nan, np.nan, np.nan
+
+        # 整合信息.
+        self.vlm_info_dict[group_name] = {}
+        self.vlm_info_dict[group_name]['time_list'] = date_list
+        self.vlm_info_dict[group_name]['value_list'] = df_g[field + '_mean'].values.tolist()
+        self.vlm_info_dict[group_name]['count_list'] = df_g[field + '_count'].values.tolist()
+        self.vlm_info_dict[group_name]['miss_rate_list'] = df_g[field + '_miss_rate'].values.tolist()
+        self.vlm_info_dict[group_name]['zero_rate_list'] = df_g[field + '_zero_rate'].values.tolist()
+
+        self.vlm_info_dict[group_name]['mean'] = mean
+        self.vlm_info_dict[group_name]['std'] = std
+        self.vlm_info_dict[group_name]['trend'] = [trend_0, trend_1, trend_2, trend_3]
+        self.vlm_info_dict[group_name]['h'] = [h_0, h_1, h_2, h_3]
+        self.vlm_info_dict[group_name]['p'] = [p_0, p_1, p_2, p_3]
+        if any([x > mean + 3 * std or x < mean - 3 * std for x in self.vlm_info_dict[group_name]['value_list']]):
+            self.vlm_info_dict[group_name]['over_3std'] = True
+        else:
+            self.vlm_info_dict[group_name]['over_3std'] = False
+
+    def process_data(self, field):
+        '''
+        对每个字段对应的不同客群进行数据处理.
+        :param field: str, 需要处理的字段.
+        :return: .
+        '''
+        # 过滤空跑数据.
+        df_copy = self.merge_data[['applied_at', 'applied_type', 'applied_channel', field]].copy()
+        df_copy = self.filter_data(df_copy, field)
+        if df_copy.shape[0] == 0:
+            print('%s还在空跑.' % self.field_query_name_dict[field])
+            return None
+        # 申请类型过滤.
+        tmp_df = pd.DataFrame()
+
+        for i in self.field_query_app_type_dict[field]:
+            print(i)
+            tmp_df = tmp_df.append(df_copy[df_copy['applied_type'] == int(i)])
+        df_copy = tmp_df
+        # 收集覆盖客群.
+        user_group_dict = {}  # user_group_dict = {'首申-融360': (1, 融360)}
+        main_user_group_dict = {}
+        app_type_dict = {1: '首申', 2: '复申', 3: '复贷'}
+        df_copy_g = df_copy.groupby(['applied_type', 'applied_channel'])[field].count().sort_values(ascending=False)
+        df_copy_g = df_copy_g.reset_index()
+        df_copy_g = df_copy_g.loc[df_copy_g[field] > 100]  # 太小的客群直接过滤.
+
+        for i in range(df_copy_g.shape[0]):
+            app_type = df_copy_g.iloc[i]['applied_type']
+            app_chan = df_copy_g.iloc[i]['applied_channel']
+            user_group_dict[app_type_dict[app_type] + '-' + app_chan] = (app_type, app_chan)
+            if df_copy_g.iloc[i][field] > int(self.min_user_group * self.num_day / 30):
+                main_user_group_dict[app_type_dict[app_type] + '-' + app_chan] = (app_type, app_chan)
+                print(app_type, app_chan)
+
+        del df_copy_g
+        # 过滤非覆盖数据.
+        tmp_df = pd.DataFrame()
+        for user_group_name in user_group_dict:
+            tmp_df = tmp_df.append(df_copy.loc[
+                              (df_copy['applied_type'] == user_group_dict[user_group_name][0]) & (
+                                      df_copy['applied_channel'] == user_group_dict[user_group_name][1])])
+        df_copy = tmp_df
+
+        ## 覆盖全样本.
+        self.process_data_helper(group_name=self.field_query_name_dict[field] + '-全样本', df=df_copy, field=field)
+        ## 首申/复申/复贷 若覆盖的话.
+        if df_copy.loc[df_copy['applied_type'] == 1].shape[0] > int(self.min_user_group * self.num_day / 30):
+            self.process_data_helper(group_name=self.field_query_name_dict[field] + '-首申',
+                                     df=df_copy.loc[df_copy['applied_type'] == 1],
+                                     field=field)
+        if df_copy.loc[df_copy['applied_type'] == 2].shape[0] > int(self.min_user_group * self.num_day / 30):
+            self.process_data_helper(group_name=self.field_query_name_dict[field] + '-复申',
+                                     df=df_copy.loc[df_copy['applied_type'] == 2],
+                                     field=field)
+        if df_copy.loc[df_copy['applied_type'] == 3].shape[0] > int(self.min_user_group * self.num_day / 30):
+            self.process_data_helper(group_name=self.field_query_name_dict[field] + '-复贷',
+                                     df=df_copy.loc[df_copy['applied_type'] == 3],
+                                     field=field)
+
+        ## 按划分的客群处理数据.
+        for user_group_name in main_user_group_dict:
+            self.process_data_helper(group_name=self.field_query_name_dict[field] + '-' + user_group_name,
+                                     df=df_copy.loc[
+                                         (df_copy['applied_type'] == main_user_group_dict[user_group_name][0]) & (
+                                                     df_copy['applied_channel'] ==
+                                                     main_user_group_dict[user_group_name][1])],
+                                     field=field)
+
+    def plot(self):
+        '''
+        根据vlm_info_dict进行绘图.
+        '''
+        for user_group_name in self.vlm_info_dict:
+            print(user_group_name)
+            plt.figure(figsize=(30, 15))
+            # 画主线.
+            plt.plot(range(self.num_day),
+                     self.vlm_info_dict[user_group_name]['value_list'])
+            # 打印趋势信息.
+            plt.text(x=0.9, y=0.75, s='整体趋势: %s\n'
+                                      '%d月趋势: %s\n'
+                                      '%d月趋势: %s\n'
+                                      '%d月趋势: %s\n' % (self.vlm_info_dict[user_group_name]['trend'][0],
+                                                       self.first_month,
+                                                       self.vlm_info_dict[user_group_name]['trend'][1],
+                                                       self.second_month,
+                                                       self.vlm_info_dict[user_group_name]['trend'][2],
+                                                       self.second_month + 1,
+                                                       self.vlm_info_dict[user_group_name]['trend'][3]),
+                     fontsize=15, transform=plt.gca().transAxes)
+
+            # 画均值, 标准差.
+            plt.hlines(y=self.vlm_info_dict[user_group_name]['mean'], xmin=0, xmax=self.num_day - 1, colors='k', linestyles='--')
+            plt.hlines(y=self.vlm_info_dict[user_group_name]['mean'] + 3 * self.vlm_info_dict[user_group_name]['std'], xmin=0,
+                       xmax=self.num_day - 1, colors='r',
+                       linestyles='--')
+            plt.hlines(y=self.vlm_info_dict[user_group_name]['mean'] - 3 * self.vlm_info_dict[user_group_name]['std'], xmin=0,
+                       xmax=self.num_day - 1, colors='r',
+                       linestyles='--')
+
+            # 画表格.
+
+            cell_text = [[str(round(x, 3)) for x in self.vlm_info_dict[user_group_name]['value_list']],
+                         [str(int(x)) for x in self.vlm_info_dict[user_group_name]['count_list']],
+                         [str(100 * x)[:4] + '%' for x in self.vlm_info_dict[user_group_name]['miss_rate_list']],
+                         [str(100 * x)[:4] + '%' for x in self.vlm_info_dict[user_group_name]['zero_rate_list']]]
+            rows = ['value', 'count', 'miss_rate', 'zero_rate']
+            cols = [x[5:] for x in self.vlm_info_dict[user_group_name]['time_list']]
+            plt.table(cellText=cell_text,
+                      rowLabels=rows,
+                      colLabels=cols,
+                      colWidths=[0.91 / (self.num_day - 1)] * self.num_day,
+                      loc='bottom')
+            plt.subplots_adjust(left=.1, bottom=.15)
+
+            # 画曲线下阴影线.
+            plt.vlines(x=range(self.num_day),
+                       ymin=[self.vlm_info_dict[user_group_name]['mean'] - 3.5 * self.vlm_info_dict[user_group_name]['std']] * self.num_day,
+                       ymax=self.vlm_info_dict[user_group_name]['value_list'],
+                       colors='lightgrey',
+                       linestyles='--')
+
+            # 展示.
+            plt.title(user_group_name + '-mean')
+            plt.grid()
+            plt.xticks([])
+
+            # 分开保存.
+            is_save = False
+            if self.vlm_info_dict[user_group_name]['h'][0] == True:
+                plt.savefig(self.fig_save_path + 'trend/' + user_group_name + '-mean')
+                is_save = True
+            if self.vlm_info_dict[user_group_name]['over_3std'] == True:
+                plt.savefig(self.fig_save_path + 'over_3std/' + user_group_name + '-mean')
+                is_save = True
+            if not is_save:
+                plt.savefig(self.fig_save_path + 'image/' + user_group_name + '-mean')
+            plt.show()
+
+    def save_vlm_info(self):
+        self.vlm_info_df = pd.DataFrame(columns=['group_name',
+                                                 'trend_0', 'trend_1', 'trend_2', 'trend_3',
+                                                 'h_0', 'h_1', 'h_2', 'h_3',
+                                                 'p_0', 'p_1', 'p_2', 'p_3',
+                                                 'over_3std'])
+        for g in self.vlm_info_dict:
+            tmp_df = pd.DataFrame({'group_name': [g],
+                                   'trend_0': [self.vlm_info_dict[g]['trend'][0]],
+                                   'trend_1': [self.vlm_info_dict[g]['trend'][1]],
+                                   'trend_2': [self.vlm_info_dict[g]['trend'][2]],
+                                   'trend_3': [self.vlm_info_dict[g]['trend'][3]],
+                                   'h_0': [self.vlm_info_dict[g]['h'][0]],
+                                   'h_1': [self.vlm_info_dict[g]['h'][1]],
+                                   'h_2': [self.vlm_info_dict[g]['h'][2]],
+                                   'h_3': [self.vlm_info_dict[g]['h'][3]],
+                                   'p_0': [self.vlm_info_dict[g]['p'][0]],
+                                   'p_1': [self.vlm_info_dict[g]['p'][1]],
+                                   'p_2': [self.vlm_info_dict[g]['p'][2]],
+                                   'p_3': [self.vlm_info_dict[g]['p'][3]],
+                                   'over_3std': [self.vlm_info_dict[g]['over_3std']]
+                                   })
+            self.vlm_info_df = self.vlm_info_df.append(tmp_df)
+        self.vlm_info_df.to_csv(self.info_save_path + 'vlm_info.csv', index=False)
+
+        with open(self.info_save_path + 'vlm_info.dict', 'wb') as f:
+            pickle.dump(self.vlm_info_dict, f)
+            f.close()
+
+    def run(self):
+        # 读取数据.
+        if self.if_read:
+            self.mysql_df = self.query_mysql('''SELECT order_no, applied_at,
+                                               applied_from, applied_channel, passdue_day
+                                               FROM risk_analysis
+                                               WHERE applied_at >= "%s 00:00:00"
+                                               AND applied_at <= "%s 00:00:00"'''
+                                             % (self.first_date, datetime.date.today().strftime('%Y-%m-%d')))
+            if self.if_save:
+                self.mysql_df.to_csv(self.data_save_path + 'mysql_data.csv', index=False)
+        else:
+            self.mysql_df = pd.read_csv(self.data_save_path + 'mysql_data.csv')
+        print('MySQL数据获取成功.')
+
+        def func_0(data):
+            try:
+                return int(int(data) + 1)
+            except:
+                return np.nan
+        if self.if_read:
+            condition = {'wf_created_at': {'$gte': '%s 00:00:00' % self.first_date,
+                                           '$lte': '%s 00:00:00' % datetime.date.today().strftime('%Y-%m-%d')}}
+            fields = {'wf_biz_no': 1, 'wf_created_at': 1, 'wf_loan_type': 1}
+            for f in self.field_query_list:  # 加入Excel中预置的模型分名称
+                fields[f] = 1
+            self.mongo_df = self.query_mongo(condition, fields)
+
+            self.mongo_df['applied_type'] = self.mongo_df['wf_loan_type'].apply(func_0)
+
+            del self.mongo_df['wf_loan_type']
+            if self.if_save:
+                self.mongo_df.to_csv(self.data_save_path + 'mongo_data.csv', index=False)
+        else:
+            self.mongo_df = pd.read_csv(self.data_save_path + 'mongo_data.csv')
+        self.mongo_df = self.mongo_df.loc[self.mongo_df['applied_type'].notna()]
+
+        print('Mongo数据获取成功.')
+
+        # MySQL数据去重.
+        self.mysql_df = self.mysql_df.sort_values('passdue_day')
+        self.mysql_df = self.mysql_df.drop_duplicates('order_no', keep='first')
+        print('数据去重完成.')
+
+        # 拼接数据.
+        self.merge_data = pd.merge(left=self.mysql_df, right=self.mongo_df,
+                                   left_on='order_no', right_on='wf_biz_no',
+                                   how='inner')
+        print('拼接数据成功')
+
+        # 清洗数据.
+        def clean_data(data):
+            try:
+                return float(data)
+            except:
+                return np.nan
+
+        na_field_list = []
+        for field in self.field_query_list:
+            if field in self.merge_data.columns.tolist():
+                print('正在清洗%s' % self.field_query_name_dict[field])
+                self.merge_data[field] = self.merge_data[field].apply(clean_data)
+            else:
+                na_field_list.append(field)
+        # 清洗时间格式, 使其转换成统一的字符串格式.
+        if repr(self.merge_data['applied_at'].dtype) == "dtype('O')":
+            self.merge_data['applied_at'] = self.merge_data['applied_at'].apply(lambda x: x[:10])
+        else:
+            self.merge_data['applied_at'] = self.merge_data['applied_at'].apply(lambda x: x.strftime('%Y-%m-%d'))
+
+        # 确认数据时间范围.
+        self.merge_data = self.merge_data.loc[
+            (self.merge_data['applied_at'] >= self.first_date) & (self.merge_data['applied_at'] <= self.current_date)]
+
+        # 去除因为一些原因未抽取到的字段.
+        print('不包含以下字段:')
+        for field in na_field_list:
+            print(self.field_query_name_dict[field])
+            self.field_query_list.remove(field)
+            self.field_name_list.remove(self.field_query_name_dict[field])
+            del self.field_query_name_dict[field]
+
+        # 处理数据.
+        print('开始处理数据.')
+        print('=' * 40)
+        for field in self.field_query_list:
+            self.process_data(field)
+        print('数据处理完毕.')
+        print('=' * 40)
+
+        # 画图.
+        print('开始画图.')
+        print('=' * 40)
+        self.plot()
+        print('画图完成.')
+        print('=' * 40)
+
+        # 保存信息.
+        print('开始保存信息.')
+        print('=' * 40)
+        self.save_vlm_info()
+        print('保存信息完成.')
+        print('=' * 40)
+
+
+if __name__ == '__main__':
+    pass

report/MM_report_20190728.html

-<!doctype html>
-<html>
-<head>
-<meta charset='UTF-8'><meta name='viewport' content='width=device-width initial-scale=1'>
-<title>MM_report_20190728</title><link href='https://fonts.loli.net/css?family=PT+Serif:400,400italic,700,700italic&subset=latin,cyrillic-ext,cyrillic,latin-ext' rel='stylesheet' type='text/css' /><style type='text/css'>html {overflow-x: initial !important;}:root { --bg-color:#ffffff; --text-color:#333333; --select-text-bg-color:#B5D6FC; --select-text-font-color:auto; --monospace:"Lucida Console",Consolas,"Courier",monospace; }
-html { font-size: 14px; background-color: var(--bg-color); color: var(--text-color); font-family: "Helvetica Neue", Helvetica, Arial, sans-serif; -webkit-font-smoothing: antialiased; }
-body { margin: 0px; padding: 0px; height: auto; bottom: 0px; top: 0px; left: 0px; right: 0px; font-size: 1rem; line-height: 1.42857; overflow-x: hidden; background: inherit; tab-size: 4; }
-iframe { margin: auto; }
-a.url { word-break: break-all; }
-a:active, a:hover { outline: 0px; }
-.in-text-selection, ::selection { text-shadow: none; background: var(--select-text-bg-color); color: var(--select-text-font-color); }
-#write { margin: 0px auto; height: auto; width: inherit; word-break: normal; word-wrap: break-word; position: relative; white-space: normal; overflow-x: visible; padding-top: 40px; }
-#write.first-line-indent p { text-indent: 2em; }
-#write.first-line-indent li p, #write.first-line-indent p * { text-indent: 0px; }
-#write.first-line-indent li { margin-left: 2em; }
-.for-image #write { padding-left: 8px; padding-right: 8px; }
-body.typora-export { padding-left: 30px; padding-right: 30px; }
-.typora-export .footnote-line, .typora-export li, .typora-export p { white-space: pre-wrap; }
-@media screen and (max-width: 500px) {
-  body.typora-export { padding-left: 0px; padding-right: 0px; }
-  #write { padding-left: 20px; padding-right: 20px; }
-  .CodeMirror-sizer { margin-left: 0px !important; }
-  .CodeMirror-gutters { display: none !important; }
-}
-#write li > figure:first-child { margin-top: -20px; }
-#write ol, #write ul { position: relative; }
-img { max-width: 100%; vertical-align: middle; }
-button, input, select, textarea { color: inherit; font-style: inherit; font-variant: inherit; font-weight: inherit; font-stretch: inherit; font-size: inherit; line-height: inherit; font-family: inherit; }
-input[type="checkbox"], input[type="radio"] { line-height: normal; padding: 0px; }
-*, ::after, ::before { box-sizing: border-box; }
-#write h1, #write h2, #write h3, #write h4, #write h5, #write h6, #write p, #write pre { width: inherit; }
-#write h1, #write h2, #write h3, #write h4, #write h5, #write h6, #write p { position: relative; }
-h1, h2, h3, h4, h5, h6 { break-after: avoid-page; break-inside: avoid; orphans: 2; }
-p { orphans: 4; }
-h1 { font-size: 2rem; }
-h2 { font-size: 1.8rem; }
-h3 { font-size: 1.6rem; }
-h4 { font-size: 1.4rem; }
-h5 { font-size: 1.2rem; }
-h6 { font-size: 1rem; }
-.md-math-block, .md-rawblock, h1, h2, h3, h4, h5, h6, p { margin-top: 1rem; margin-bottom: 1rem; }
-.hidden { display: none; }
-.md-blockmeta { color: rgb(204, 204, 204); font-weight: 700; font-style: italic; }
-a { cursor: pointer; }
-sup.md-footnote { padding: 2px 4px; background-color: rgba(238, 238, 238, 0.7); color: rgb(85, 85, 85); border-radius: 4px; cursor: pointer; }
-sup.md-footnote a, sup.md-footnote a:hover { color: inherit; text-transform: inherit; text-decoration: inherit; }
-#write input[type="checkbox"] { cursor: pointer; width: inherit; height: inherit; }
-figure { overflow-x: auto; margin: 1.2em 0px; max-width: calc(100% + 16px); padding: 0px; }
-figure > table { margin: 0px !important; }
-tr { break-inside: avoid; break-after: auto; }
-thead { display: table-header-group; }
-table { border-collapse: collapse; border-spacing: 0px; width: 100%; overflow: auto; break-inside: auto; text-align: left; }
-table.md-table td { min-width: 32px; }
-.CodeMirror-gutters { border-right: 0px; background-color: inherit; }
-.CodeMirror-linenumber { user-select: none; }
-.CodeMirror { text-align: left; }
-.CodeMirror-placeholder { opacity: 0.3; }
-.CodeMirror pre { padding: 0px 4px; }
-.CodeMirror-lines { padding: 0px; }
-div.hr:focus { cursor: none; }
-#write pre { white-space: pre-wrap; }
-#write.fences-no-line-wrapping pre { white-space: pre; }
-#write pre.ty-contain-cm { white-space: normal; }
-.CodeMirror-gutters { margin-right: 4px; }
-.md-fences { font-size: 0.9rem; display: block; break-inside: avoid; text-align: left; overflow: visible; white-space: pre; background: inherit; position: relative !important; }
-.md-diagram-panel { width: 100%; margin-top: 10px; text-align: center; padding-top: 0px; padding-bottom: 8px; overflow-x: auto; }
-#write .md-fences.mock-cm { white-space: pre-wrap; }
-.md-fences.md-fences-with-lineno { padding-left: 0px; }
-#write.fences-no-line-wrapping .md-fences.mock-cm { white-space: pre; overflow-x: auto; }
-.md-fences.mock-cm.md-fences-with-lineno { padding-left: 8px; }
-.CodeMirror-line, twitterwidget { break-inside: avoid; }
-.footnotes { opacity: 0.8; font-size: 0.9rem; margin-top: 1em; margin-bottom: 1em; }
-.footnotes + .footnotes { margin-top: 0px; }
-.md-reset { margin: 0px; padding: 0px; border: 0px; outline: 0px; vertical-align: top; background: 0px 0px; text-decoration: none; text-shadow: none; float: none; position: static; width: auto; height: auto; white-space: nowrap; cursor: inherit; -webkit-tap-highlight-color: transparent; line-height: normal; font-weight: 400; text-align: left; box-sizing: content-box; direction: ltr; }
-li div { padding-top: 0px; }
-blockquote { margin: 1rem 0px; }
-li .mathjax-block, li p { margin: 0.5rem 0px; }
-li { margin: 0px; position: relative; }
-blockquote > :last-child { margin-bottom: 0px; }
-blockquote > :first-child, li > :first-child { margin-top: 0px; }
-.footnotes-area { color: rgb(136, 136, 136); margin-top: 0.714rem; padding-bottom: 0.143rem; white-space: normal; }
-#write .footnote-line { white-space: pre-wrap; }
-@media print {
-  body, html { border: 1px solid transparent; height: 99%; break-after: avoid; break-before: avoid; }
-  #write { margin-top: 0px; padding-top: 0px; border-color: transparent !important; }
-  .typora-export * { -webkit-print-color-adjust: exact; }
-  html.blink-to-pdf { font-size: 13px; }
-  .typora-export #write { padding-left: 32px; padding-right: 32px; padding-bottom: 0px; break-after: avoid; }
-  .typora-export #write::after { height: 0px; }
-  @page { margin: 20mm 0px; }
-}
-.footnote-line { margin-top: 0.714em; font-size: 0.7em; }
-a img, img a { cursor: pointer; }
-pre.md-meta-block { font-size: 0.8rem; min-height: 0.8rem; white-space: pre-wrap; background: rgb(204, 204, 204); display: block; overflow-x: hidden; }
-p > .md-image:only-child:not(.md-img-error) img, p > img:only-child { display: block; margin: auto; }
-p > .md-image:only-child { display: inline-block; width: 100%; }
-#write .MathJax_Display { margin: 0.8em 0px 0px; }
-.md-math-block { width: 100%; }
-.md-math-block:not(:empty)::after { display: none; }
-[contenteditable="true"]:active, [contenteditable="true"]:focus { outline: 0px; box-shadow: none; }
-.md-task-list-item { position: relative; list-style-type: none; }
-.task-list-item.md-task-list-item { padding-left: 0px; }
-.md-task-list-item > input { position: absolute; top: 0px; left: 0px; margin-left: -1.2em; margin-top: calc(1em - 10px); border: none; }
-.math { font-size: 1rem; }
-.md-toc { min-height: 3.58rem; position: relative; font-size: 0.9rem; border-radius: 10px; }
-.md-toc-content { position: relative; margin-left: 0px; }
-.md-toc-content::after, .md-toc::after { display: none; }
-.md-toc-item { display: block; color: rgb(65, 131, 196); }
-.md-toc-item a { text-decoration: none; }
-.md-toc-inner:hover { text-decoration: underline; }
-.md-toc-inner { display: inline-block; cursor: pointer; }
-.md-toc-h1 .md-toc-inner { margin-left: 0px; font-weight: 700; }
-.md-toc-h2 .md-toc-inner { margin-left: 2em; }
-.md-toc-h3 .md-toc-inner { margin-left: 4em; }
-.md-toc-h4 .md-toc-inner { margin-left: 6em; }
-.md-toc-h5 .md-toc-inner { margin-left: 8em; }
-.md-toc-h6 .md-toc-inner { margin-left: 10em; }
-@media screen and (max-width: 48em) {
-  .md-toc-h3 .md-toc-inner { margin-left: 3.5em; }
-  .md-toc-h4 .md-toc-inner { margin-left: 5em; }
-  .md-toc-h5 .md-toc-inner { margin-left: 6.5em; }
-  .md-toc-h6 .md-toc-inner { margin-left: 8em; }
-}
-a.md-toc-inner { font-size: inherit; font-style: inherit; font-weight: inherit; line-height: inherit; }
-.footnote-line a:not(.reversefootnote) { color: inherit; }
-.md-attr { display: none; }
-.md-fn-count::after { content: "."; }
-code, pre, samp, tt { font-family: var(--monospace); }
-kbd { margin: 0px 0.1em; padding: 0.1em 0.6em; font-size: 0.8em; color: rgb(36, 39, 41); background: rgb(255, 255, 255); border: 1px solid rgb(173, 179, 185); border-radius: 3px; box-shadow: rgba(12, 13, 14, 0.2) 0px 1px 0px, rgb(255, 255, 255) 0px 0px 0px 2px inset; white-space: nowrap; vertical-align: middle; }
-.md-comment { color: rgb(162, 127, 3); opacity: 0.8; font-family: var(--monospace); }
-code { text-align: left; vertical-align: initial; }
-a.md-print-anchor { white-space: pre !important; border-width: initial !important; border-style: none !important; border-color: initial !important; display: inline-block !important; position: absolute !important; width: 1px !important; right: 0px !important; outline: 0px !important; background: 0px 0px !important; text-decoration: initial !important; text-shadow: initial !important; }
-.md-inline-math .MathJax_SVG .noError { display: none !important; }
-.html-for-mac .inline-math-svg .MathJax_SVG { vertical-align: 0.2px; }
-.md-math-block .MathJax_SVG_Display { text-align: center; margin: 0px; position: relative; text-indent: 0px; max-width: none; max-height: none; min-height: 0px; min-width: 100%; width: auto; overflow-y: hidden; display: block !important; }
-.MathJax_SVG_Display, .md-inline-math .MathJax_SVG_Display { width: auto; margin: inherit; display: inline-block !important; }
-.MathJax_SVG .MJX-monospace { font-family: var(--monospace); }
-.MathJax_SVG .MJX-sans-serif { font-family: sans-serif; }
-.MathJax_SVG { display: inline; font-style: normal; font-weight: 400; line-height: normal; zoom: 90%; text-indent: 0px; text-align: left; text-transform: none; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; direction: ltr; max-width: none; max-height: none; min-width: 0px; min-height: 0px; border: 0px; padding: 0px; margin: 0px; }
-.MathJax_SVG * { transition: none; }
-.MathJax_SVG_Display svg { vertical-align: middle !important; margin-bottom: 0px !important; }
-.os-windows.monocolor-emoji .md-emoji { font-family: "Segoe UI Symbol", sans-serif; }
-.md-diagram-panel > svg { max-width: 100%; }
-[lang="mermaid"] svg, [lang="flow"] svg { max-width: 100%; }
-[lang="mermaid"] .node text { font-size: 1rem; }
-table tr th { border-bottom: 0px; }
-video { max-width: 100%; display: block; margin: 0px auto; }
-iframe { max-width: 100%; width: 100%; border: none; }
-.highlight td, .highlight tr { border: 0px; }
-
-
-:root { --active-file-bg-color: rgba(32, 43, 51, 0.63); --active-file-text-color: white; --bg-color: #f3f2ee; --text-color: #1f0909; --control-text-color: #444; --rawblock-edit-panel-bd: #e5e5e5; --select-text-bg-color: rgba(32, 43, 51, 0.63); --select-text-font-color: white; }
-pre { --select-text-bg-color: #36284e; --select-text-font-color: #fff; }
-html { font-size: 16px; }
-html, body { background-color: rgb(243, 242, 238); font-family: "PT Serif", "Times New Roman", Times, serif; color: rgb(31, 9, 9); line-height: 1.5em; }
-#write { max-width: 40em; }
-ol li { list-style-type: decimal; list-style-position: outside; }
-ul li { list-style-type: disc; list-style-position: outside; }
-ol, ul { list-style: none; }
-blockquote, q { quotes: none; }
-blockquote::before, blockquote::after, q::before, q::after { content: none; }
-table { border-collapse: collapse; border-spacing: 0px; }
-h1, h2, h3, h4, h5, h6 { font-weight: bold; }
-h1 { font-size: 1.875em; line-height: 1.6em; margin-top: 2em; }
-h2, h3 { font-size: 1.3125em; line-height: 1.15; margin-top: 2.28571em; margin-bottom: 1.15em; }
-h3 { font-weight: normal; }
-h4 { font-size: 1.125em; margin-top: 2.67em; }
-h5, h6 { font-size: 1em; }
-h1 { border-bottom: 1px solid; margin-bottom: 1.875em; padding-bottom: 0.8125em; }
-a { text-decoration: none; color: rgb(6, 85, 136); }
-a:hover, a:active { text-decoration: underline; }
-p, blockquote, .md-fences { margin-bottom: 1.5em; }
-h1, h2, h3, h4, h5, h6 { margin-bottom: 1.5em; }
-blockquote { font-style: italic; border-left: 5px solid; margin-left: 2em; padding-left: 1em; }
-ul, ol { margin: 0px 0px 1.5em 1.5em; }
-.md-meta, .md-before, .md-after { color: rgb(153, 153, 153); }
-table { margin-bottom: 1.5em; font-size: 1em; }
-thead th, tfoot th { padding: 0.25em 0.25em 0.25em 0.4em; text-transform: uppercase; }
-th { text-align: left; }
-td { vertical-align: top; padding: 0.25em 0.25em 0.25em 0.4em; }
-code, .md-fences { background-color: rgb(218, 218, 218); }
-code { padding-left: 2px; padding-right: 2px; }
-.md-fences { margin-left: 2em; margin-bottom: 3em; padding-left: 1ch; padding-right: 1ch; }
-pre, code, tt { font-size: 0.875em; line-height: 1.71429em; }
-h1 { line-height: 1.3em; font-weight: normal; margin-bottom: 0.5em; }
-p + ul, p + ol { margin-top: 0.5em; }
-h3 + ul, h4 + ul, h5 + ul, h6 + ul, h3 + ol, h4 + ol, h5 + ol, h6 + ol { margin-top: 0.5em; }
-li > ul, li > ol { margin-top: inherit; margin-bottom: 0px; }
-li ol > li { list-style-type: lower-alpha; }
-li li ol > li { list-style-type: lower-roman; }
-h2, h3 { margin-bottom: 0.75em; }
-hr { border-top: none; border-right: none; border-bottom: 1px solid; border-left: none; }
-h1 { border-color: rgb(197, 197, 197); }
-blockquote { border-color: rgb(186, 186, 186); color: rgb(101, 101, 101); }
-blockquote ul, blockquote ol { margin-left: 0px; }
-.ty-table-edit { background-color: transparent; }
-thead { background-color: rgb(218, 218, 218); }
-tr:nth-child(2n) { background: rgb(232, 231, 231); }
-hr { border-color: rgb(197, 197, 197); }
-.task-list { padding-left: 1rem; }
-.md-task-list-item { padding-left: 1.5rem; list-style-type: none; }
-.md-task-list-item > input::before { content: "√"; display: inline-block; width: 1.25rem; height: 1.6rem; vertical-align: middle; text-align: center; color: rgb(221, 221, 221); background-color: rgb(243, 242, 238); }
-.md-task-list-item > input:checked::before, .md-task-list-item > input[checked]::before { color: inherit; }
-#write pre.md-meta-block { min-height: 1.875rem; color: rgb(85, 85, 85); border: 0px; background: transparent; margin-left: 1em; margin-top: 1em; }
-.md-image > .md-meta { color: rgb(155, 81, 70); }
-.md-image > .md-meta { font-family: Menlo, "Ubuntu Mono", Consolas, "Courier New", "Microsoft Yahei", "Hiragino Sans GB", "WenQuanYi Micro Hei", serif; }
-#write > h3.md-focus::before { left: -1.5rem; color: rgb(153, 153, 153); border-color: rgb(153, 153, 153); }
-#write > h4.md-focus::before { left: -1.5rem; top: 0.25rem; color: rgb(153, 153, 153); border-color: rgb(153, 153, 153); }
-#write > h5.md-focus::before { left: -1.5rem; color: rgb(153, 153, 153); border-color: rgb(153, 153, 153); }
-#write > h6.md-focus::before { left: -1.5rem; top: 0.3125rem; color: rgb(153, 153, 153); border-color: rgb(153, 153, 153); }
-.md-toc:focus .md-toc-content { margin-top: 19px; }
-.md-toc-content:empty::before { color: rgb(6, 85, 136); }
-.md-toc-item { color: rgb(6, 85, 136); }
-#write div.md-toc-tooltip { background-color: rgb(243, 242, 238); }
-#typora-sidebar { background-color: rgb(243, 242, 238); box-shadow: rgba(0, 0, 0, 0.376) 0px 6px 12px; }
-.pin-outline #typora-sidebar { background: inherit; box-shadow: none; border-right: 1px dashed; }
-.pin-outline #typora-sidebar:hover .outline-title-wrapper { border-left: 1px dashed; }
-.outline-item:hover { background-color: rgb(218, 218, 218); border-left: 28px solid rgb(218, 218, 218); border-right: 18px solid rgb(218, 218, 218); }
-.typora-node .outline-item:hover { border-right: 28px solid rgb(218, 218, 218); }
-.outline-expander::before { content: ""; font-family: FontAwesome; font-size: 14px; top: 1px; }
-.outline-expander:hover::before, .outline-item-open > .outline-item > .outline-expander::before { content: ""; }
-.modal-content { background-color: rgb(243, 242, 238); }
-.auto-suggest-container ul li { list-style-type: none; }
-.megamenu-menu, #top-titlebar, #top-titlebar *, .megamenu-content { background: rgb(243, 242, 238); color: rgb(31, 9, 9); }
-.megamenu-menu-header { border-bottom: 1px dashed rgb(32, 43, 51); }
-.megamenu-menu { box-shadow: none; border-right: 1px dashed; }
-header, .context-menu, .megamenu-content, footer { font-family: "PT Serif", "Times New Roman", Times, serif; color: rgb(31, 9, 9); }
-#megamenu-back-btn { color: rgb(31, 9, 9); border-color: rgb(31, 9, 9); }
-.megamenu-menu-header #megamenu-menu-header-title::before { color: rgb(31, 9, 9); }
-.megamenu-menu-list li a:hover, .megamenu-menu-list li a.active { color: inherit; background-color: rgb(232, 231, 223); }
-.long-btn:hover { background-color: rgb(232, 231, 223); }
-#recent-file-panel tbody tr:nth-child(2n-1) { background-color: transparent !important; }
-.megamenu-menu-panel tbody tr:hover td:nth-child(2) { color: inherit; }
-.megamenu-menu-panel .btn { background-color: rgb(210, 209, 209); }
-.btn-default { background-color: transparent; }
-.typora-sourceview-on #toggle-sourceview-btn, .ty-show-word-count #footer-word-count { background: rgb(199, 197, 197); }
-#typora-quick-open { background-color: inherit; }
-.md-diagram-panel { margin-top: 8px; }
-.file-list-item-file-name { font-weight: initial; }
-.file-list-item-summary { opacity: 1; }
-.file-list-item { color: rgb(119, 119, 119); }
-.file-list-item.active { background-color: inherit; color: black; }
-.ty-side-sort-btn.active { background-color: inherit; }
-.file-list-item.active .file-list-item-file-name { font-weight: bold; }
-.file-list-item { opacity: 1 !important; }
-.file-library-node.active > .file-node-background { background-color: var(--active-file-bg-color); }
-.file-tree-node.active > .file-node-content { color: var(--active-file-text-color); }
-.md-task-list-item > input { margin-left: -1.6em; margin-top: calc(1rem - 12px); }
-input { border: 1px solid rgb(170, 170, 170); }
-.megamenu-menu-header #megamenu-menu-header-title, .megamenu-menu-header:hover, .megamenu-menu-header:focus { color: inherit; }
-.dropdown-menu .divider { border-color: rgb(229, 229, 229); }
-.os-windows-7 strong, .os-windows-7 strong { font-weight: 760; }
-
-
-#write { counter-reset: h1 0; }
-h1 { counter-reset: h2 0; }
-h2 { counter-reset: h3 0; }
-h3 { counter-reset: h4 0; }
-h4 { counter-reset: h5 0; }
-h5 { counter-reset: h6 0; }
-#write h1::before { counter-increment: h1 1; content: counter(h1) " "; }
-#write h2::before { counter-increment: h2 1; content: counter(h1) "." counter(h2) " "; }
-#write h3::before, h3.md-focus.md-heading::before { counter-increment: h3 1; content: counter(h1) "." counter(h2) "." counter(h3) " "; }
-#write h4::before, h4.md-focus.md-heading::before { counter-increment: h4 1; content: counter(h1) "." counter(h2) "." counter(h3) "." counter(h4) " "; }
-#write h5::before, h5.md-focus.md-heading::before { counter-increment: h5 1; content: counter(h1) "." counter(h2) "." counter(h3) "." counter(h4) "." counter(h5) " "; }
-#write h6::before, h6.md-focus.md-heading::before { counter-increment: h6 1; content: counter(h1) "." counter(h2) "." counter(h3) "." counter(h4) "." counter(h5) "." counter(h6) " "; }
-#write > h3.md-focus::before, #write > h4.md-focus::before, #write > h5.md-focus::before, #write > h6.md-focus::before, h3.md-focus::before, h4.md-focus::before, h5.md-focus::before, h6.md-focus::before { color: inherit; border: inherit; border-radius: inherit; position: inherit; left: initial; float: none; top: initial; font-size: inherit; padding-left: inherit; padding-right: inherit; vertical-align: inherit; font-weight: inherit; line-height: inherit; }
-
-
-
- .typora-export li, .typora-export p, .typora-export,  .footnote-line {white-space: normal;} 
-</style>
-</head>
-<body class='typora-export os-windows' >
-<div  id='write'  class = 'is-node'><div class='md-toc' mdtype='toc'><p class="md-toc-content"><span class="md-toc-item md-toc-h1" data-ref="n2"><a class="md-toc-inner" href="#header-n2">概述</a></span><span class="md-toc-item md-toc-h2" data-ref="n3"><a class="md-toc-inner" href="#header-n3">报告内容</a></span><span class="md-toc-item md-toc-h2" data-ref="n11"><a class="md-toc-inner" href="#header-n11">监控的模型</a></span><span class="md-toc-item md-toc-h2" data-ref="n64"><a class="md-toc-inner" href="#header-n64">时间跨度</a></span><span class="md-toc-item md-toc-h1" data-ref="n68"><a class="md-toc-inner" href="#header-n68">PSI</a></span><span class="md-toc-item md-toc-h2" data-ref="n69"><a class="md-toc-inner" href="#header-n69">与整体差别较大的客群</a></span><span class="md-toc-item md-toc-h3" data-ref="n70"><a class="md-toc-inner" href="#header-n70">概述</a></span><span class="md-toc-item md-toc-h3" data-ref="n78"><a class="md-toc-inner" href="#header-n78">有哪些客群</a></span><span class="md-toc-item md-toc-h3" data-ref="n144"><a class="md-toc-inner" href="#header-n144">资质相对好的客群</a></span><span class="md-toc-item md-toc-h3" data-ref="n153"><a class="md-toc-inner" href="#header-n153">资质相对差的客群</a></span><span class="md-toc-item md-toc-h2" data-ref="n158"><a class="md-toc-inner" href="#header-n158">随时间分布变化较大的客群</a></span><span class="md-toc-item md-toc-h3" data-ref="n159"><a class="md-toc-inner" href="#header-n159">有哪些客群</a></span><span class="md-toc-item md-toc-h3" data-ref="n200"><a class="md-toc-inner" href="#header-n200">资质变好的客群</a></span><span class="md-toc-item md-toc-h3" data-ref="n214"><a class="md-toc-inner" href="#header-n214">资质变差的客群</a></span><span class="md-toc-item md-toc-h1" data-ref="n226"><a class="md-toc-inner" href="#header-n226">AUC</a></span><span class="md-toc-item md-toc-h2" data-ref="n227"><a class="md-toc-inner" href="#header-n227">概述</a></span><span class="md-toc-item md-toc-h2" data-ref="n233"><a class="md-toc-inner" href="#header-n233">相比整体模型效果较差的客群</a></span><span class="md-toc-item md-toc-h2" data-ref="n255"><a class="md-toc-inner" href="#header-n255">模型效果随时间变差的客群</a></span></p></div><h1><a name='header-n2' class='md-header-anchor '></a>概述</h1><h2><a name='header-n3' class='md-header-anchor '></a>报告内容</h2><ul><li>本次模型报告由于代码处于开发阶段, 所以内容并不完整.</li><li>主要包括模型分的PSI, 以及AUC随时间的变化, 以及客群与整体的差别.</li><li>尽管模型监控代码加入了自动筛选异常客群的功能, 但要更加深入分析仍需要时间, 此版报告为简略版.</li></ul><h2><a name='header-n11' class='md-header-anchor '></a>监控的模型</h2><ul><li><p>已监控模型:</p><ul><li>同盾V1</li><li>V6</li><li>百融V1</li><li>首贷融合模型</li><li>融360首贷融合模型</li><li>新颜V2</li><li>新颜V3</li><li>魔蝎</li><li>复贷新融合模型(无运营商)</li><li>复贷新融合模型(有运营商)</li><li>短信模型</li><li>探知</li><li>同盾V2</li><li>电话邦V2</li><li>百融V2</li><li>复贷V4</li><li>首贷新融合模型(无运营商)</li><li>首贷新融合模型(有运营商)</li></ul></li><li><p>暂未监控模型:</p><ul><li><p>不明原因(待与模型负责人交流):</p><ul><li>APP模型</li><li>量信分V2</li></ul></li><li><p>还有一些模型待加入.</p></li></ul></li></ul><h2><a name='header-n64' class='md-header-anchor '></a>时间跨度</h2><ul><li>考虑到整体模型的时间跨度, 本次模型监控报告的取数时间范围为2019.03.01 - 2019.07. 26.</li></ul><h1><a name='header-n68' class='md-header-anchor '></a>PSI</h1><h2><a name='header-n69' class='md-header-anchor '></a>与整体差别较大的客群</h2><h3><a name='header-n70' class='md-header-anchor '></a>概述</h3><ul><li>由模型监控代码在已监控模型中, 产生的有效客群数量为465个.</li><li>设定PSI阈值(0.5)后, 筛选出的与整体差别较大的异常客群有15个.</li><li>设定PSI阈值(0.1)后, 筛选出的变化较大的异常客群有25个.</li></ul><h3><a name='header-n78' class='md-header-anchor '></a>有哪些客群</h3><ul><li><p>客群列举:</p><figure><table><thead><tr><th style='text-align:center;' >客群名称</th><th style='text-align:center;' >月份</th><th style='text-align:center;' >PSI</th></tr></thead><tbody><tr><td style='text-align:center;' >同盾V1-复贷-百融榕树</td><td style='text-align:center;' >6</td><td style='text-align:center;' >0.57</td></tr><tr><td style='text-align:center;' >V6-首申-国美api</td><td style='text-align:center;' >6</td><td style='text-align:center;' >2.47</td></tr><tr><td style='text-align:center;' >V6-首申-壹账通H5</td><td style='text-align:center;' >3</td><td style='text-align:center;' >1.75</td></tr><tr><td style='text-align:center;' >V6-首申-平安一贷通安卓</td><td style='text-align:center;' >5</td><td style='text-align:center;' >2.07</td></tr><tr><td style='text-align:center;' >V6-复贷-51公积金API</td><td style='text-align:center;' >3</td><td style='text-align:center;' >0.53</td></tr><tr><td style='text-align:center;' >新颜V2-复申-挖财api</td><td style='text-align:center;' >4</td><td style='text-align:center;' >0.57</td></tr><tr><td style='text-align:center;' >首贷融合模型-复贷-全渠道</td><td style='text-align:center;' >5</td><td style='text-align:center;' >1.09</td></tr><tr><td style='text-align:center;' >首贷融合模型-首申-壹账通H5</td><td style='text-align:center;' >3</td><td style='text-align:center;' >1.89</td></tr><tr><td style='text-align:center;' >首贷融合模型-首申-平安一贷通安卓</td><td style='text-align:center;' >5</td><td style='text-align:center;' >2.13</td></tr><tr><td style='text-align:center;' >首贷融合模型-复申-挖财api</td><td style='text-align:center;' >4</td><td style='text-align:center;' >0.51</td></tr><tr><td style='text-align:center;' >融360首贷融合模型-复贷-全渠道</td><td style='text-align:center;' >5</td><td style='text-align:center;' >1.01</td></tr><tr><td style='text-align:center;' >融360首贷融合模型-首申-壹账通H5</td><td style='text-align:center;' >3</td><td style='text-align:center;' >1.82</td></tr><tr><td style='text-align:center;' >融360首贷融合模型-首申-平安一贷通安卓</td><td style='text-align:center;' >5</td><td style='text-align:center;' >2.13</td></tr><tr><td style='text-align:center;' >融360首贷融合模型-复申-挖财api</td><td style='text-align:center;' >4</td><td style='text-align:center;' >0.67</td></tr></tbody></table></figure><p>&nbsp;</p></li></ul><h3><a name='header-n144' class='md-header-anchor '></a>资质相对好的客群</h3><ul><li><p>暂时鉴别方式为在异常的基础上去看图, 准备后续加入趋势判断, 自动识别是变好还是变差.</p></li><li><p>经过查看, 均为资质相对较好的客群.</p></li><li><p>例如:</p><p><img src='../doc/image/融360首贷融合模型-复申-挖财api.png' alt='融360首贷融合模型-复申-挖财api' referrerPolicy='no-referrer' /></p></li></ul><h3><a name='header-n153' class='md-header-anchor '></a>资质相对差的客群</h3><ul><li>无.</li></ul><p>&nbsp;</p><h2><a name='header-n158' class='md-header-anchor '></a>随时间分布变化较大的客群</h2><h3><a name='header-n159' class='md-header-anchor '></a>有哪些客群</h3><ul><li><p>部分客群列举:</p><figure><table><thead><tr><th style='text-align:center;' >客群名称</th><th style='text-align:center;' >月份</th><th style='text-align:center;' >PSI</th></tr></thead><tbody><tr><td style='text-align:center;' >同盾分v1-首申-百融榕树</td><td style='text-align:center;' >5</td><td style='text-align:center;' >0.29</td></tr><tr><td style='text-align:center;' >V6-全样本</td><td style='text-align:center;' >7</td><td style='text-align:center;' >0.10</td></tr><tr><td style='text-align:center;' >新颜V2-复贷-全渠道</td><td style='text-align:center;' >5</td><td style='text-align:center;' >0.66</td></tr><tr><td style='text-align:center;' >新颜v3-全样本</td><td style='text-align:center;' >6</td><td style='text-align:center;' >0.10</td></tr><tr><td style='text-align:center;' >魔蝎-全样本</td><td style='text-align:center;' >6</td><td style='text-align:center;' >0.34</td></tr><tr><td style='text-align:center;' >短信分-全样本</td><td style='text-align:center;' >7</td><td style='text-align:center;' >0.13</td></tr><tr><td style='text-align:center;' >融360首贷融合模型-首申-融360</td><td style='text-align:center;' >4</td><td style='text-align:center;' >0.12</td></tr><tr><td style='text-align:center;' >电话邦V2-首申-全渠道</td><td style='text-align:center;' >6</td><td style='text-align:center;' >0.13</td></tr></tbody></table></figure></li></ul><h3><a name='header-n200' class='md-header-anchor '></a>资质变好的客群</h3><ul><li><p>暂时鉴别方式为在异常的基础上去看图, 准备后续加入自动识别.</p></li><li><p>部分资质变好客群列举:</p><figure><table><thead><tr><th style='text-align:center;' >客群名称</th></tr></thead><tbody><tr><td style='text-align:center;' >V6-全样本</td></tr><tr><td style='text-align:center;' >新颜V2-复贷-全渠道</td></tr></tbody></table></figure><p><img src='../doc/image/V6-全样本.png' alt='V6-全样本' referrerPolicy='no-referrer' /></p></li></ul><h3><a name='header-n214' class='md-header-anchor '></a>资质变差的客群</h3><ul><li><p>部分资质变差客群列举:</p><figure><table><thead><tr><th style='text-align:center;' >客群名称</th></tr></thead><tbody><tr><td style='text-align:center;' >同盾分v1-首申-百融榕树</td></tr><tr><td style='text-align:center;' >短信分-全样本</td></tr></tbody></table></figure><p><img src='../doc/image/同盾分v1-首申-百融榕树.png' alt='同盾分v1-首申-百融榕树' referrerPolicy='no-referrer' /></p></li></ul><h1><a name='header-n226' class='md-header-anchor '></a>AUC</h1><h2><a name='header-n227' class='md-header-anchor '></a>概述</h2><ul><li>由代码产生的客群为113个, 因为本身放款数量相对申请数量少, 且需要响应时间.</li><li>设定AUCR阈值为0.85, 得到的异常客群数量为23个.</li></ul><h2><a name='header-n233' class='md-header-anchor '></a>相比整体模型效果较差的客群</h2><ul><li><p>客群列举:</p><figure><table><thead><tr><th style='text-align:center;' >客群名称</th><th style='text-align:center;' >月份</th><th style='text-align:center;' >AUCR</th></tr></thead><tbody><tr><td style='text-align:center;' >新颜V2-首申-挖财api</td><td style='text-align:center;' >4</td><td style='text-align:center;' >0.80</td></tr><tr><td style='text-align:center;' >新颜V2-首申-爱奇艺H5</td><td style='text-align:center;' >3</td><td style='text-align:center;' >0.80</td></tr><tr><td style='text-align:center;' >新颜v3-首申-360金融API</td><td style='text-align:center;' >4</td><td style='text-align:center;' >0.85</td></tr></tbody></table></figure><p><img src='../doc/image/新颜V2-首申-爱奇艺H5.png' alt='新颜V2-首申-爱奇艺H5' referrerPolicy='no-referrer' /></p></li></ul><h2><a name='header-n255' class='md-header-anchor '></a>模型效果随时间变差的客群</h2><ul><li><p>部分客群列举:</p><figure><table><thead><tr><th style='text-align:center;' >客群名称</th><th style='text-align:center;' >月份</th><th style='text-align:center;' >AUCR</th></tr></thead><tbody><tr><td style='text-align:center;' >同盾分v1-首申-爱奇艺H5</td><td style='text-align:center;' >4</td><td style='text-align:center;' >0.76</td></tr><tr><td style='text-align:center;' >同盾分v1-首申-挖财api</td><td style='text-align:center;' >6</td><td style='text-align:center;' >0.77</td></tr><tr><td style='text-align:center;' >V6-复申-全渠道</td><td style='text-align:center;' >6</td><td style='text-align:center;' >0.70</td></tr><tr><td style='text-align:center;' >首贷融合模型-首申-融360</td><td style='text-align:center;' >5</td><td style='text-align:center;' >0.75</td></tr></tbody></table></figure><p><img src='../doc/image/V6-复申-全渠道.png' alt='V6-复申-全渠道' referrerPolicy='no-referrer' /></p></li></ul><p>&nbsp;</p></div>
-</body>
-</html>
\ No newline at end of file

report/MM_report_20190728.md

-[TOC]
-
-# 概述
-
-## 报告内容
-
-- 本次模型报告由于代码处于开发阶段, 所以内容并不完整.
-- 主要包括模型分的PSI, 以及AUC随时间的变化, 以及客群与整体的差别.
-- 尽管模型监控代码加入了自动筛选异常客群的功能, 但要更加深入分析仍需要时间, 此版报告为简略版.
-
-## 监控的模型
-
-- 已监控模型:
-  - 同盾V1
-  - V6
-  - 百融V1
-  - 首贷融合模型
-  - 融360首贷融合模型
-  - 新颜V2
-  - 新颜V3
-  - 魔蝎
-  - 复贷新融合模型(无运营商)
-  - 复贷新融合模型(有运营商)
-  - 短信模型
-  - 探知
-  - 同盾V2
-  - 电话邦V2
-  - 百融V2
-  - 复贷V4
-  - 首贷新融合模型(无运营商)
-  - 首贷新融合模型(有运营商)
-- 暂未监控模型:
-  - 不明原因(待与模型负责人交流):
-    - APP模型
-    - 量信分V2
-  - 还有一些模型待加入.
-
-## 时间跨度
-
-- 考虑到整体模型的时间跨度, 本次模型监控报告的取数时间范围为2019.03.01 - 2019.07. 26.
-
-# PSI
-
-## 与整体差别较大的客群
-
-### 概述
-
-- 由模型监控代码在已监控模型中, 产生的有效客群数量为465个.
-- 设定PSI阈值(0.5)后, 筛选出的与整体差别较大的异常客群有15个.
-- 设定PSI阈值(0.1)后, 筛选出的变化较大的异常客群有25个.
-
-### 有哪些客群
-
-- 客群列举:
-
-  |               客群名称                | 月份 | PSI  |
-  | :-----------------------------------: | :--: | :--: |
-  |         同盾V1-复贷-百融榕树          |  6   | 0.57 |
-  |            V6-首申-国美api            |  6   | 2.47 |
-  |           V6-首申-壹账通H5            |  3   | 1.75 |
-  |        V6-首申-平安一贷通安卓         |  5   | 2.07 |
-  |          V6-复贷-51公积金API          |  3   | 0.53 |
-  |          新颜V2-复申-挖财api          |  4   | 0.57 |
-  |       首贷融合模型-复贷-全渠道        |  5   | 1.09 |
-  |      首贷融合模型-首申-壹账通H5       |  3   | 1.89 |
-  |   首贷融合模型-首申-平安一贷通安卓    |  5   | 2.13 |
-  |       首贷融合模型-复申-挖财api       |  4   | 0.51 |
-  |     融360首贷融合模型-复贷-全渠道     |  5   | 1.01 |
-  |    融360首贷融合模型-首申-壹账通H5    |  3   | 1.82 |
-  | 融360首贷融合模型-首申-平安一贷通安卓 |  5   | 2.13 |
-  |    融360首贷融合模型-复申-挖财api     |  4   | 0.67 |
-
-  
-
-### 资质相对好的客群
-
-- 暂时鉴别方式为在异常的基础上去看图, 准备后续加入趋势判断, 自动识别是变好还是变差.
-
-- 经过查看, 均为资质相对较好的客群.
-
-- 例如:
-
-  ![融360首贷融合模型-复申-挖财api](../doc/image/融360首贷融合模型-复申-挖财api.png)
-
-### 资质相对差的客群
-
-- 无.
-
-
-
-## 随时间分布变化较大的客群
-
-### 有哪些客群
-
-- 部分客群列举:
-
-  |           客群名称           | 月份 | PSI  |
-  | :--------------------------: | :--: | :--: |
-  |    同盾分v1-首申-百融榕树    |  5   | 0.29 |
-  |          V6-全样本           |  7   | 0.10 |
-  |      新颜V2-复贷-全渠道      |  5   | 0.66 |
-  |        新颜v3-全样本         |  6   | 0.10 |
-  |         魔蝎-全样本          |  6   | 0.34 |
-  |        短信分-全样本         |  7   | 0.13 |
-  | 融360首贷融合模型-首申-融360 |  4   | 0.12 |
-  |     电话邦V2-首申-全渠道     |  6   | 0.13 |
-
-### 资质变好的客群
-
-- 暂时鉴别方式为在异常的基础上去看图, 准备后续加入自动识别.
-
-- 部分资质变好客群列举:
-
-  |      客群名称      |
-  | :----------------: |
-  |     V6-全样本      |
-  | 新颜V2-复贷-全渠道 |
-
-  ![V6-全样本](../doc/image/V6-全样本.png)
-
-### 资质变差的客群
-
-- 部分资质变差客群列举:
-
-  |        客群名称        |
-  | :--------------------: |
-  | 同盾分v1-首申-百融榕树 |
-  |     短信分-全样本      |
-
-  ![同盾分v1-首申-百融榕树](../doc/image/同盾分v1-首申-百融榕树.png)
-
-# AUC
-
-## 概述
-
-- 由代码产生的客群为113个, 因为本身放款数量相对申请数量少, 且需要响应时间.
-- 设定AUCR阈值为0.85, 得到的异常客群数量为23个.
-
-## 相比整体模型效果较差的客群
-
-- 客群列举:
-
-  |        客群名称        | 月份 | AUCR |
-  | :--------------------: | :--: | :--: |
-  |  新颜V2-首申-挖财api   |  4   | 0.80 |
-  |  新颜V2-首申-爱奇艺H5  |  3   | 0.80 |
-  | 新颜v3-首申-360金融API |  4   | 0.85 |
-
-  ![新颜V2-首申-爱奇艺H5](../doc/image/新颜V2-首申-爱奇艺H5.png)
-
-## 模型效果随时间变差的客群
-
-- 部分客群列举:
-
-  |        客群名称         | 月份 | AUCR |
-  | :---------------------: | :--: | :--: |
-  | 同盾分v1-首申-爱奇艺H5  |  4   | 0.76 |
-  |  同盾分v1-首申-挖财api  |  6   | 0.77 |
-  |     V6-复申-全渠道      |  6   | 0.70 |
-  | 首贷融合模型-首申-融360 |  5   | 0.75 |
-
-  ![V6-复申-全渠道](../doc/image/V6-复申-全渠道.png)
-