In [1]:

    

import pandas as pd


    

In [2]:

    

# set measurement levels
col_types = {'chain': 'object',
             'offer': 'object',
             'market': 'object',
             'category': 'object',
             'company': 'object',
             'brand': 'object',
             'exact_item_bought': 'object'}


    

In [3]:

    

# read data created from SAS key
data_pd = pd.read_csv('/home/patrickh/workspace/GWU_data_mining/01_basic_data_prep/assignment/key/assignment_1_key_train_sas.csv', dtype=col_types)
score_pd = pd.read_csv('/home/patrickh/workspace/GWU_data_mining/01_basic_data_prep/assignment/key/assignment_1_key_test_sas.csv', dtype=col_types)


    

In [4]:

    

# confirm shape
data_pd.shape


    

    
Out[4]:





(160057, 14)

In [5]:

    

# confirm shape
score_pd.shape


    

    
Out[5]:





(151484, 13)

In [6]:

    

# data['chain'].unique().shape    # 130 levels
# data['market'].unique().shape   # 34 levels
# data['category'].unique().shape # 13 levels
# data['brand'].unique().shape    # 12 levels
# data['company'].unique().shape  # 11 levels


    

In [7]:

    

# show column names
data_pd.columns


    

    
Out[7]:





Index(['id', 'chain', 'offer', 'market', 'repeater', 'offerdate', 'category',
       'quantity', 'company', 'offervalue', 'brand', 'avg_category_quantity',
       'avg_category_amount', 'exact_item_bought'],
      dtype='object')

In [8]:

    

# ensure repeater==t is modeled as 1
data_pd.loc[data_pd['repeater'] == 't', 'repeater'] = 1
data_pd.loc[data_pd['repeater'] == 'f', 'repeater'] = 0


    

In [9]:

    

# fix unknown values in train and test
chain_unknowns =\
list(set(data_pd['chain'].unique()) - set(score_pd['chain'].unique())) +\
list(set(score_pd['chain'].unique()) - set(data_pd['chain'].unique()))
    
print(chain_unknowns)     

data_pd.loc[data_pd['chain'].isin(chain_unknowns), 'chain'] = 'unknown'
score_pd.loc[score_pd['chain'].isin(chain_unknowns), 'chain'] = 'unknown'
     
print(sorted(data_pd['chain'].unique()))
print(len(data_pd['chain'].unique()))
print(sorted(score_pd['chain'].unique()))
print(len(score_pd['chain'].unique()))


    

    




['9', '230', '369', '284', '387', '370', '421']
['10', '100', '101', '104', '106', '108', '109', '115', '116', '12', '122', '126', '133', '134', '14', '140', '143', '15', '151', '152', '153', '16', '161', '163', '165', '166', '169', '17', '18', '180', '184', '187', '191', '192', '2', '20', '205', '206', '21', '211', '214', '215', '217', '224', '23', '233', '24', '240', '241', '246', '26', '278', '285', '293', '3', '304', '306', '307', '31', '313', '356', '360', '362', '368', '373', '376', '377', '384', '386', '388', '389', '390', '391', '392', '393', '394', '395', '396', '397', '398', '399', '4', '40', '400', '401', '402', '403', '404', '405', '42', '422', '424', '431', '46', '48', '507', '508', '509', '510', '520', '521', '522', '523', '524', '525', '526', '58', '6', '62', '63', '64', '65', '68', '69', '70', '71', '73', '77', '8', '81', '83', '85', '88', '89', '95', '96', '98', 'unknown']
128
['10', '100', '101', '104', '106', '108', '109', '115', '116', '12', '122', '126', '133', '134', '14', '140', '143', '15', '151', '152', '153', '16', '161', '163', '165', '166', '169', '17', '18', '180', '184', '187', '191', '192', '2', '20', '205', '206', '21', '211', '214', '215', '217', '224', '23', '233', '24', '240', '241', '246', '26', '278', '285', '293', '3', '304', '306', '307', '31', '313', '356', '360', '362', '368', '373', '376', '377', '384', '386', '388', '389', '390', '391', '392', '393', '394', '395', '396', '397', '398', '399', '4', '40', '400', '401', '402', '403', '404', '405', '42', '422', '424', '431', '46', '48', '507', '508', '509', '510', '520', '521', '522', '523', '524', '525', '526', '58', '6', '62', '63', '64', '65', '68', '69', '70', '71', '73', '77', '8', '81', '83', '85', '88', '89', '95', '96', '98', 'unknown']
128

In [10]:

    

# fix unknown values in train and test
market_unknowns =\
list(set(data_pd['market'].unique()) - set(score_pd['market'].unique())) +\
list(set(score_pd['market'].unique()) - set(data_pd['market'].unique()))
    
print(market_unknowns)     
     
print(sorted(data_pd['market'].unique()))
print(len(data_pd['market'].unique()))
print(sorted(score_pd['market'].unique()))
print(len(score_pd['market'].unique()))


    

    




[]
['1', '10', '11', '12', '14', '15', '16', '17', '18', '2', '20', '21', '22', '23', '24', '26', '27', '28', '33', '34', '35', '37', '39', '4', '43', '45', '47', '5', '6', '7', '8', '9', '93', '96']
34
['1', '10', '11', '12', '14', '15', '16', '17', '18', '2', '20', '21', '22', '23', '24', '26', '27', '28', '33', '34', '35', '37', '39', '4', '43', '45', '47', '5', '6', '7', '8', '9', '93', '96']
34

In [11]:

    

# fix unknown values in train and test
category_unknowns =\
list(set(data_pd['category'].unique()) - set(score_pd['category'].unique())) +\
list(set(score_pd['category'].unique()) - set(data_pd['category'].unique()))
    
print(category_unknowns)     

data_pd.loc[data_pd['category'].isin(category_unknowns), 'category'] = 'unknown'
score_pd.loc[score_pd['category'].isin(category_unknowns), 'category'] = 'unknown'
     
print(sorted(data_pd['category'].unique()))
print(len(data_pd['category'].unique()))
print(sorted(score_pd['category'].unique()))
print(len(score_pd['category'].unique()))


    

    




['3509', '9909', '3203', '1703', '4401', '7205', '5824', '799', '4517', '9115', '706', '5122']
['1726', '2119', '2202', '3504', '5558', '5616', '5619', '6202', 'unknown']
9
['1726', '2119', '2202', '3504', '5558', '5616', '5619', '6202', 'unknown']
9

In [12]:

    

# fix unknown values in train and test
brand_unknowns =\
list(set(data_pd['brand'].unique()) - set(score_pd['brand'].unique())) +\
list(set(score_pd['brand'].unique()) - set(data_pd['brand'].unique()))
    
print(brand_unknowns)     

data_pd.loc[data_pd['brand'].isin(brand_unknowns), 'brand'] = 'unknown'
score_pd.loc[score_pd['brand'].isin(brand_unknowns), 'brand'] = 'unknown'
     
print(sorted(data_pd['brand'].unique()))
print(len(data_pd['brand'].unique()))
print(sorted(score_pd['brand'].unique()))
print(len(score_pd['brand'].unique()))


    

    




['13791', '6732', '875', '28840', '13474', '26456', '4294', '93904', '17286', '1322', '17311', '26189']
['102504', '15889', '3718', '5072', '64486', '6926', '7668', 'unknown']
8
['102504', '15889', '3718', '5072', '64486', '6926', '7668', 'unknown']
8

In [13]:

    

# fix unknown values in train and test
company_unknowns =\
list(set(data_pd['company'].unique()) - set(score_pd['company'].unique())) +\
list(set(score_pd['company'].unique()) - set(data_pd['company'].unique()))
    
print(company_unknowns)     
     
data_pd.loc[data_pd['company'].isin(company_unknowns), 'company'] = 'unknown'
score_pd.loc[score_pd['company'].isin(company_unknowns), 'company'] = 'unknown'    
    
print(sorted(data_pd['company'].unique()))
print(len(data_pd['company'].unique()))
print(sorted(score_pd['company'].unique()))
print(len(score_pd['company'].unique()))


    

    




['106414464', '105100050', '103320030', '1089520383', '107127979', '103700030', '105190050', '104127141', '107106878', '1076211171', '108500080', '105450050']
['104460040', '104610040', '107120272', '107717272', '108079383', '1087744888', 'unknown']
7
['104460040', '104610040', '107120272', '107717272', '108079383', '1087744888', 'unknown']
7

In [14]:

    

# fix unknown values in train and test
offer_unknowns =\
list(set(data_pd['offer'].unique()) - set(score_pd['offer'].unique())) +\
list(set(score_pd['offer'].unique()) - set(data_pd['offer'].unique()))
    
print(offer_unknowns)     

data_pd.loc[data_pd['offer'].isin(offer_unknowns), 'offer'] = 'unknown'
score_pd.loc[score_pd['offer'].isin(offer_unknowns), 'offer'] = 'unknown'    
    
print(sorted(data_pd['offer'].unique()))
print(len(data_pd['offer'].unique()))
print(sorted(score_pd['offer'].unique()))
print(len(score_pd['offer'].unique()))


    

    




['1199258', '1200578', '1197502', '1199256', '1200988', '1194044', '1200579', '1203052', '1190530', '1219903', '1221665', '1220502', '1219900', '1221663', '1221658', '1221667', '1230218', '1203439', '1221666', '1220503', '1213242']
['1198271', '1198272', '1198273', '1198274', '1198275', '1200581', '1200582', '1200584', '1204576', '1204821', '1204822', '1208251', '1208252', '1208329', '1208501', '1208503', 'unknown']
17
['1198271', '1198272', '1198273', '1198274', '1198275', '1200581', '1200582', '1200584', '1204576', '1204821', '1204822', '1208251', '1208252', '1208329', '1208501', '1208503', 'unknown']
17

In [15]:

    

# start and import h2o
# set seed
import h2o
h2o.init()
from h2o.estimators.glm import H2OGeneralizedLinearEstimator

SEED = 12345


    

    




Checking whether there is an H2O instance running at http://localhost:54321..... not found.
Attempting to start a local H2O server...
  Java Version: java version "1.8.0_201"; Java(TM) SE Runtime Environment (build 1.8.0_201-b09); Java HotSpot(TM) 64-Bit Server VM (build 25.201-b09, mixed mode)
  Starting server from /home/patrickh/anaconda3/lib/python3.6/site-packages/h2o/backend/bin/h2o.jar
  Ice root: /tmp/tmpdauqerlp
  JVM stdout: /tmp/tmpdauqerlp/h2o_patrickh_started_from_python.out
  JVM stderr: /tmp/tmpdauqerlp/h2o_patrickh_started_from_python.err
  Server is running at http://127.0.0.1:54321
Connecting to H2O server at http://127.0.0.1:54321... successful.






    





H2O cluster uptime:
00 secs
H2O cluster timezone:
America/New_York
H2O data parsing timezone:
UTC
H2O cluster version:
3.22.1.4
H2O cluster version age:
1 month and 1 day 
H2O cluster name:
H2O_from_python_patrickh_w6n0dr
H2O cluster total nodes:
1
H2O cluster free memory:
3.422 Gb
H2O cluster total cores:
8
H2O cluster allowed cores:
8
H2O cluster status:
accepting new members, healthy
H2O connection url:
http://127.0.0.1:54321
H2O connection proxy:
None
H2O internal security:
False
H2O API Extensions:
Amazon S3, XGBoost, Algos, AutoML, Core V3, Core V4
Python version:
3.6.4 final

In [16]:

    

# enforce same measurement levels in h2o and pandas
col_types = {'chain': 'enum',
             'offer': 'enum',
             'market': 'enum',
             'category': 'enum',
             'company': 'enum',
             'brand': 'enum',
             'exact_item_bought': 'enum'}

data_h2o = h2o.H2OFrame(data_pd, column_types=col_types)
score_h2o = h2o.H2OFrame(score_pd, column_types=col_types)


    

    




Parse progress: |█████████████████████████████████████████████████████████| 100%
Parse progress: |█████████████████████████████████████████████████████████| 100%

In [17]:

    

# expand date into new features
data_h2o = data_h2o.concat(data_h2o['offerdate'].month().rename({'offerdate': 'month'}).asfactor())
data_h2o = data_h2o.concat(data_h2o['offerdate'].week().rename({'offerdate': 'week'}).asfactor())
data_h2o = data_h2o.concat(data_h2o['offerdate'].dayOfWeek().rename({'offerdate': 'dayOfWeek'}))
data_h2o = data_h2o.concat(data_h2o['offerdate'].day().rename({'offerdate': 'day'}))

score_h2o = score_h2o.concat(score_h2o['offerdate'].month().rename({'offerdate': 'month'}).asfactor())
score_h2o = score_h2o.concat(score_h2o['offerdate'].week().rename({'offerdate': 'week'}).asfactor())
score_h2o = score_h2o.concat(score_h2o['offerdate'].dayOfWeek().rename({'offerdate': 'dayOfWeek'}))
score_h2o = score_h2o.concat(score_h2o['offerdate'].day().rename({'offerdate': 'day'}))


    

In [18]:

    

# look at training data
data_h2o.describe()


    

    




Rows:160057
Cols:18








    







       
id                
chain  
offer  
market  
repeater           
offerdate          
category  
quantity  
company  
offervalue        
brand  
avg_category_quantity  
avg_category_amount  
exact_item_bought  
month  
week  
dayOfWeek  
day               


type   
int               
enum   
enum   
enum    
int                
time               
enum      
int       
enum     
real              
enum   
real                   
real                 
enum               
enum   
enum  
enum       
int               
mins   
86246.0           
       
       
        
0.0                
1362096000000.0    
          
1.0       
         
0.75              
       
-1.0                   
-85.14               
                   
       
      
           
1.0               
mean   
1869324131.2555754
       
       
        
0.27139081702143614
1365220653082.3323 
          
1.0       
         
1.25532310364433  
       
0.6949999031818261     
2.2441312757227134   
                   
       
      
           
19.004467158574705
maxs   
4809911000.0      
       
       
        
1.0                
1367280000000.0    
          
1.0       
         
3.0               
       
44.181818182           
62.218               
                   
       
      
           
31.0              
sigma  
1570833451.4856217
       
       
        
0.4446786220260453 
1303790212.0504372 
          
0.0       
         
0.5246445261465219
       
0.7609072934172032     
2.5989523363669176   
                   
       
      
           
9.632905002480289 
zeros  
0                 
       
       
        
116619             
0                  
          
0         
         
0                 
       
72549                  
72591                
                   
       
      
           
0                 
missing
0                 
0      
0      
0       
0                  
0                  
0         
0         
0        
0                 
0      
0                      
0                    
0                  
0      
0     
0          
0                 
0      
86246.0           
205    
1208251
34      
1.0                
2013-04-24 00:00:00
2202      
1.0       
104460040
2.0               
3718   
0.0                    
0.0                  
0                  
4      
17    
Wed        
24.0              
1      
86252.0           
205    
unknown
34      
1.0                
2013-03-27 00:00:00
unknown   
1.0       
unknown  
0.75              
unknown
1.0                    
2.425                
1                  
3      
13    
Wed        
27.0              
2      
12682470.0        
18     
unknown
11      
0.0                
2013-03-28 00:00:00
unknown   
1.0       
unknown  
0.75              
unknown
1.0                    
2.5                  
1                  
3      
13    
Thu        
28.0              
3      
12996040.0        
15     
unknown
9       
0.0                
2013-03-25 00:00:00
unknown   
1.0       
unknown  
0.75              
unknown
0.0                    
0.0                  
0                  
3      
13    
Mon        
25.0              
4      
13089312.0        
15     
1204821
9       
0.0                
2013-04-01 00:00:00
5619      
1.0       
107717272
1.5               
102504 
0.0                    
0.0                  
0                  
4      
14    
Mon        
1.0               
5      
13179265.0        
14     
unknown
8       
0.0                
2013-03-29 00:00:00
unknown   
1.0       
unknown  
0.75              
unknown
0.0                    
0.0                  
0                  
3      
13    
Fri        
29.0              
6      
13251776.0        
15     
1200581
9       
0.0                
2013-03-30 00:00:00
1726      
1.0       
104460040
1.25              
7668   
1.8                    
4.563                
1                  
3      
13    
Sat        
30.0              
7      
13540129.0        
14     
1200581
8       
0.0                
2013-03-30 00:00:00
1726      
1.0       
104460040
1.25              
7668   
1.3333333333           
4.6533333333         
1                  
3      
13    
Sat        
30.0              
8      
13807224.0        
4      
1204576
1       
0.0                
2013-04-05 00:00:00
5616      
1.0       
104610040
1.0               
15889  
1.4375                 
3.365                
1                  
4      
14    
Fri        
5.0               
9      
13873775.0        
4      
unknown
1       
0.0                
2013-03-26 00:00:00
unknown   
1.0       
unknown  
0.75              
unknown
0.0                    
0.0                  
0                  
3      
13    
Tue        
26.0              

In [19]:

    

# look at test data
score_h2o.describe()


    

    




Rows:151484
Cols:17








    







       
id                
chain  
offer  
market  
offerdate          
category  
quantity          
company  
offervalue        
brand  
avg_category_quantity  
avg_category_amount  
exact_item_bought  
month  
week  
dayOfWeek  
day               


type   
int               
enum   
enum   
enum    
time               
enum      
int               
enum     
real              
enum   
real                   
real                 
enum               
enum   
enum  
enum       
int               
mins   
12262064.0        
       
       
        
1367366400000.0    
          
1.0               
         
1.0               
       
-1.0                   
-18.76               
                   
       
      
           
1.0               
mean   
2367235598.062943 
       
       
        
1372045742758.312  
          
1.2141942383354016
         
1.8039594940719834
       
0.9798367012155557     
3.918267425267764    
                   
       
      
           
20.815993768318656
maxs   
4853598737.0      
       
       
        
1375228800000.0    
          
2.0               
         
5.0               
       
40.0                   
103.87               
                   
       
      
           
31.0              
sigma  
1629644569.9344237
       
       
        
1768808429.4717584 
          
0.4102635466557285
         
0.7684280524816689
       
0.6520652050938551     
3.24780201201996     
                   
       
      
           
7.076741234531553 
zeros  
0                 
       
       
        
0                  
          
0                 
         
0                 
       
27950                  
27970                
                   
       
      
           
0                 
missing
0                 
0      
0      
0       
0                  
0         
0                 
0        
0                 
0      
0                      
0                    
0                  
0      
0     
0          
0                 
0      
12262064.0        
95     
unknown
39      
2013-06-27 00:00:00
unknown   
1.0               
unknown  
1.5               
unknown
0.0                    
0.0                  
0                  
6      
26    
Thu        
27.0              
1      
12277270.0        
95     
unknown
39      
2013-06-23 00:00:00
unknown   
2.0               
unknown  
3.0               
unknown
1.2                    
3.882                
1                  
6      
25    
Sun        
23.0              
2      
12332190.0        
95     
unknown
39      
2013-06-15 00:00:00
unknown   
1.0               
unknown  
2.0               
unknown
2.0                    
6.98                 
1                  
6      
24    
Sat        
15.0              
3      
12524696.0        
4      
unknown
1       
2013-06-20 00:00:00
unknown   
1.0               
unknown  
1.5               
unknown
1.0                    
4.2425               
1                  
6      
25    
Thu        
20.0              
4      
13074629.0        
14     
unknown
8       
2013-06-21 00:00:00
unknown   
2.0               
unknown  
3.0               
unknown
1.0                    
5.6185714286         
1                  
6      
25    
Fri        
21.0              
5      
13387341.0        
14     
unknown
8       
2013-06-22 00:00:00
unknown   
2.0               
unknown  
3.0               
unknown
1.0                    
4.49                 
1                  
6      
25    
Sat        
22.0              
6      
13501141.0        
4      
unknown
1       
2013-05-13 00:00:00
unknown   
1.0               
unknown  
1.5               
unknown
1.0                    
6.74                 
1                  
5      
20    
Mon        
13.0              
7      
13558712.0        
15     
unknown
9       
2013-05-12 00:00:00
unknown   
1.0               
unknown  
1.5               
unknown
1.0                    
10.323333333         
1                  
5      
19    
Sun        
12.0              
8      
13563017.0        
14     
unknown
8       
2013-06-24 00:00:00
unknown   
1.0               
unknown  
1.5               
unknown
1.0                    
5.3066666667         
1                  
6      
26    
Mon        
24.0              
9      
13584134.0        
14     
unknown
8       
2013-06-27 00:00:00
unknown   
1.0               
unknown  
1.5               
unknown
1.0                    
2.84                 
1                  
6      
26    
Thu        
27.0              

In [20]:

    

# quantity unary, drop it
# drop other unusable variables
# set modeling roles
drops = ['id', 'chain', 'market', 'offerdate', 'quantity']
y = 'repeater'
X = [name for name in data_h2o.columns if name not in [y] + drops]
print(y)
print(X)


    

    




repeater
['offer', 'category', 'company', 'offervalue', 'brand', 'avg_category_quantity', 'avg_category_amount', 'exact_item_bought', 'month', 'week', 'dayOfWeek', 'day']

In [21]:

    

# create modeling partitions
train, valid, test = data_h2o.split_frame([0.4, 0.3], seed=SEED)


    

In [22]:

    

# check shape 
train.shape


    

    
Out[22]:





(64122, 18)

In [23]:

    

# check shape 
valid.shape


    

    
Out[23]:





(47980, 18)

In [24]:

    

# check shape 
test.shape


    

    
Out[24]:





(47955, 18)

In [25]:

    

# elastic net regularized regression 
#   - L1 for variable selection
#   - L2 for handling multicollinearity
#   - IRLS for handling outliers
#   - with lamba parameter tuning for variable selection

# initialize
rptr_glm = H2OGeneralizedLinearEstimator(family='binomial',
                                         model_id='rptr_glm1',
                                         solver='IRLSM',
                                         nfolds=3,
                                         standardize=True,
                                         seed=SEED,
                                         lambda_search=True)

# train 
rptr_glm.train(X, y, training_frame=train, validation_frame=valid)


    

    




glm Model Build progress: |███████████████████████████████████████████████| 100%

In [26]:

    

# check for stability across folds -- looks good
rptr_glm.cross_validation_metrics_summary().as_data_frame()


    

    
Out[26]:







  

    

      

      

      
mean
      
sd
      
cv_1_valid
      
cv_2_valid
      
cv_3_valid
    
  
  

    

      
0
      
accuracy
      
0.5633181
      
0.0034985114
      
0.5576324
      
0.5626291
      
0.56969273
    
    

      
1
      
auc
      
0.66687614
      
0.0013872668
      
0.6641128
      
0.6684734
      
0.6680421
    
    

      
2
      
err
      
0.43668193
      
0.0034985114
      
0.4423676
      
0.4373709
      
0.4303073
    
    

      
3
      
err_count
      
9334.0
      
99.13627
      
9514.0
      
9316.0
      
9172.0
    
    

      
4
      
f0point5
      
0.39562488
      
8.234713E-4
      
0.39596537
      
0.3968501
      
0.39405915
    
    

      
5
      
f1
      
0.48101103
      
0.0012617736
      
0.48254105
      
0.481984
      
0.47850809
    
    

      
6
      
f2
      
0.613405
      
0.0024687524
      
0.61756927
      
0.6136203
      
0.60902536
    
    

      
7
      
lift_top_group
      
2.0142968
      
0.050682575
      
2.027856
      
2.0945127
      
1.920521
    
    

      
8
      
logloss
      
0.5482143
      
0.0020346912
      
0.5511301
      
0.5492147
      
0.54429805
    
    

      
9
      
max_per_class_error
      
0.50603294
      
0.0070085404
      
0.5175562
      
0.5071829
      
0.4933597
    
    

      
10
      
mcc
      
0.21990056
      
0.0016726145
      
0.21801764
      
0.21844749
      
0.22323658
    
    

      
11
      
mean_per_class_accuracy
      
0.6226167
      
0.0014614518
      
0.6208214
      
0.62151676
      
0.6255119
    
    

      
12
      
mean_per_class_error
      
0.37738332
      
0.0014614518
      
0.37917858
      
0.37848327
      
0.37448812
    
    

      
13
      
mse
      
0.18361354
      
9.3339913E-4
      
0.18492535
      
0.18410787
      
0.18180738
    
    

      
14
      
null_deviance
      
24907.797
      
143.848
      
25161.13
      
24899.21
      
24663.049
    
    

      
15
      
precision
      
0.3537607
      
7.134559E-4
      
0.35366338
      
0.3550422
      
0.35257646
    
    

      
16
      
r2
      
0.0670119
      
9.0821454E-4
      
0.06541767
      
0.068562925
      
0.067055106
    
    

      
17
      
recall
      
0.7512663
      
0.004308973
      
0.759199
      
0.75021636
      
0.7443835
    
    

      
18
      
residual_deviance
      
23435.428
      
146.4661
      
23706.31
      
23396.547
      
23203.426
    
    

      
19
      
rmse
      
0.42849872
      
0.0010901511
      
0.43002948
      
0.42907792
      
0.42638877
    
    

      
20
      
specificity
      
0.49396706
      
0.0070085404
      
0.4824438
      
0.4928171
      
0.50664026
    
  

In [27]:

    

# train AUC
rptr_glm.auc(valid=False)


    

    
Out[27]:





0.6682898515349129

In [28]:

    

# valid AUC
rptr_glm.auc(valid=True)


    

    
Out[28]:





0.6642797159900581

In [29]:

    

# test AUC
rptr_glm.model_performance(test).auc()


    

    
Out[29]:





0.66930388703468

In [30]:

    

# many validation metrics
rptr_glm.model_performance(valid)


    

    




ModelMetricsBinomialGLM: glm
** Reported on test data. **

MSE: 0.18433338418488365
RMSE: 0.4293406388695154
LogLoss: 0.5500303205580226
Null degrees of freedom: 47979
Residual degrees of freedom: 47926
Null deviance: 56034.26930760575
Residual deviance: 52780.909560747896
AIC: 52888.909560747896
AUC: 0.6642797159900581
pr_auc: 0.4082413348072177
Gini: 0.32855943198011617
Confusion Matrix (Act/Pred) for max f1 @ threshold = 0.23680137514237834: 






    






0
1
Error
Rate
0
17240.0
17753.0
0.5073
 (17753.0/34993.0)
1
3305.0
9682.0
0.2545
 (3305.0/12987.0)
Total
20545.0
27435.0
0.4389
 (21058.0/47980.0)






    




Maximum Metrics: Maximum metrics at their respective thresholds







    





metric
threshold
value
idx
max f1
0.2368014
0.4790461
274.0
max f2
0.1223229
0.6605821
358.0
max f0point5
0.3284445
0.4251892
202.0
max accuracy
0.4805090
0.7320342
76.0
max precision
0.7387865
0.6923077
8.0
max recall
0.0559133
1.0
399.0
max specificity
0.9378370
0.9999143
0.0
max absolute_mcc
0.2928417
0.2189525
229.0
max min_per_class_accuracy
0.2913873
0.6202098
231.0
max mean_per_class_accuracy
0.2928417
0.6222223
229.0






    




Gains/Lift Table: Avg response rate: 27.07 %, avg score: 26.98 %







    






group
cumulative_data_fraction
lower_threshold
lift
cumulative_lift
response_rate
score
cumulative_response_rate
cumulative_score
capture_rate
cumulative_capture_rate
gain
cumulative_gain

1
0.0100459
0.5366650
2.0005291
2.0005291
0.5414938
0.5850936
0.5414938
0.5850936
0.0200970
0.0200970
100.0529096
100.0529096

2
0.0200917
0.5128554
1.9698696
1.9851994
0.5331950
0.5228512
0.5373444
0.5539724
0.0197890
0.0398860
96.9869646
98.5199371

3
0.0300125
0.4961243
1.9093237
1.9601182
0.5168067
0.5044425
0.5305556
0.5376000
0.0189420
0.0588281
90.9323674
96.0118238

4
0.0400167
0.4826037
1.9472902
1.9569112
0.5270833
0.4896555
0.5296875
0.5256139
0.0194810
0.0783091
94.7290239
95.6911238

5
0.05
0.4676137
1.7353953
1.9126819
0.4697286
0.4742862
0.5177157
0.5153654
0.0173250
0.0956341
73.5395264
91.2681913

6
0.1
0.4247933
1.7386617
1.8256718
0.4706128
0.4440226
0.4941642
0.4796940
0.0869331
0.1825672
73.8661739
82.5671826

7
0.15
0.3940127
1.4676215
1.7063217
0.3972489
0.4091581
0.4618591
0.4561820
0.0733811
0.2559483
46.7621468
70.6321706

8
0.2
0.3653559
1.3043813
1.6058366
0.3530638
0.3791261
0.4346603
0.4369180
0.0652191
0.3211673
30.4381304
60.5836606

9
0.3152980
0.3233892
1.3456877
1.5107055
0.3642444
0.3402462
0.4089106
0.4015671
0.1551552
0.4763225
34.5687698
51.0705474

10
0.4043560
0.3039213
1.1542497
1.4321974
0.3124269
0.3130243
0.3876604
0.3820659
0.1027951
0.5791176
15.4249715
43.2197384

11
0.5002293
0.2727063
0.9581511
1.3413423
0.2593478
0.2891334
0.3630682
0.3642546
0.0918611
0.6709787
-4.1848872
34.1342304

12
0.6016048
0.2177024
0.9608340
1.2772234
0.2600740
0.2480176
0.3457128
0.3446676
0.0974051
0.7683838
-3.9166001
27.7223392

13
0.7013964
0.1838723
0.6767011
1.1917837
0.1831662
0.1940466
0.3225864
0.3232379
0.0675291
0.8359128
-32.3298943
19.1783730

14
0.8050855
0.1746898
0.6928512
1.1275249
0.1875377
0.1783366
0.3051931
0.3045757
0.0718411
0.9077539
-30.7148819
12.7524917

15
0.9000208
0.1270697
0.6277750
1.0748106
0.1699232
0.1564416
0.2909247
0.2889503
0.0595981
0.9673520
-37.2225050
7.4810629

16
1.0
0.0519661
0.3265484
1.0
0.0883886
0.0972660
0.2706753
0.2697859
0.0326480
1.0
-67.3451614
0.0






    










    
Out[30]:

In [31]:

    

# print coefficients
for key_ in rptr_glm.coef():
    print(key_, rptr_glm.coef()[key_])


    

    




Intercept 0.9142362343476567
offer.1198271 0.03463083385893831
offer.1198272 0.14052728909397258
offer.1198273 -0.028110797822582217
offer.1198274 -0.13117283777912048
offer.1198275 -0.07475893905600814
offer.1200581 -0.2663110424214166
offer.1200582 0.0
offer.1200584 0.0
offer.1204576 -0.13863085089106314
offer.1204821 -0.16653161796482882
offer.1204822 0.0
offer.1208251 0.0
offer.1208252 0.90290635746174
offer.1208329 0.03988082619178288
offer.1208501 0.4323275425518459
offer.1208503 0.0
offer.unknown -0.24095314385202235
week.9 0.3882866662473721
week.10 0.34503199800881224
week.11 -0.3374664786715979
week.12 0.0
week.13 -0.3946215550940864
week.14 -0.22160719365547848
week.15 0.0
week.16 -0.09617188647612415
week.17 0.053510178998351796
week.18 0.15565259807093573
category.1726 -0.16996755533119035
category.2119 0.03988082619178368
category.2202 0.6589080081306594
category.3504 0.0
category.5558 -0.04144651740120281
category.5616 -0.13863085089106691
category.5619 -0.09783190861526495
category.6202 0.12060177946446535
category.unknown -0.2409531438519952
brand.102504 -0.09783190861526239
brand.15889 -0.13863085089106675
brand.3718 0.6589080081306663
brand.5072 -0.04144651740120234
brand.64486 0.12060177946446629
brand.6926 0.03988082619178345
brand.7668 -0.9124249815547996
brand.unknown 0.0
company.104460040 0.0
company.104610040 -0.1386308508910667
company.107120272 -0.041446517401202365
company.107717272 -0.09783190861526334
company.108079383 0.03988082619178309
company.1087744888 0.1206017794644651
company.unknown 0.0
dayOfWeek.Mon -0.06788130486673358
dayOfWeek.Tue -0.001410333705654479
dayOfWeek.Wed 0.007728177083505018
dayOfWeek.Thu 0.0
dayOfWeek.Fri 0.0
dayOfWeek.Sat 0.14530381687259547
dayOfWeek.Sun 0.1032571462037334
exact_item_bought.0 -0.1507968420390149
exact_item_bought.1 0.1263809799828914
month.3 0.19741069489701107
month.4 -0.1549780733058436
offervalue -0.9912946388035481
avg_category_quantity -0.03619275098428959
avg_category_amount 0.08680975561720691
day -0.03191424771152614

In [32]:

    

# find id == 13584134
score_h2o['id'].asfactor().head()


    

    







      id


12262064
12277270
12332190
12524696
13074629
13387341
13501141
13558712
13563017
13584134








    
Out[32]:

In [33]:

    

# get probability for id == 13584134
rptr_glm.predict(score_h2o)


    

    




glm prediction progress: |████████████████████████████████████████████████| 100%






    




/home/patrickh/anaconda3/lib/python3.6/site-packages/h2o/job.py:69: UserWarning: Test/Validation dataset column 'week' has levels not trained on: [19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31]
  warnings.warn(w)
/home/patrickh/anaconda3/lib/python3.6/site-packages/h2o/job.py:69: UserWarning: Test/Validation dataset column 'month' has levels not trained on: [5, 6, 7]
  warnings.warn(w)






    







  predict
      p0
       p1


        0
0.931928
0.0680718
        0
0.964497
0.0355034
        0
0.854873
0.145127 
        0
0.856192
0.143808 
        0
0.960205
0.0397945
        0
0.959612
0.0403879
        0
0.804033
0.195967 
        1
0.71045 
0.28955  
        0
0.868433
0.131567 
        0
0.893706
0.106294 








    
Out[33]:

In [ ]: