features

APPLIFT


yes:
PublisherId
AppSiteId
TrafficType
AppSiteCategory
Position
OS
DeviceType
Country
CreativeType


maybe:
Age
Gender
BidFloor??????
OSVersion???
Carrier???


output:
ExchangeBid


Outcome: matters...
0
1
2
	

	
'TrafficType','AppSiteCategory','Position','Age','Gender','OS','DeviceType','Country','CreativeType'
	
	model = graphlab.logistic_classifier.create(products2,target='Outcome',features=['TrafficType','AppSiteCategory','Position','Age','Gender','OS','DeviceType','Country','CreativeType'],validation_set=test_data)
	
	
	
	import graphlab
	products=graphlab.SFrame('ustrain.csv')
	products1=products[products['Age']!=0]
	len(products1)
	products2=products1[products1['Gender']!=0]
	len(products2)
	products2.show()
	train_data,test_data = products2.random_split(.8)
	model=graphlab.logistic_classifier.create(products2,target='Outcome',features=['TrafficType','AppSiteCategory','Position','Age','Gender','OS','DeviceType','Country','CreativeType'],validation_set=test_data)
	
	Op logs:
	PROGRESS: WARNING: Detected extremely low variance for feature(s) 'Country' because all entries are nearly the same.
Proceeding with model training using all features. If the model does not provide results of adequate quality, exclude the above mentioned feature(s) from the input dataset.
PROGRESS: Logistic regression:
PROGRESS: --------------------------------------------------------
PROGRESS: Number of examples          : 2577279
PROGRESS: Number of classes           : 3
PROGRESS: Number of feature columns   : 9
PROGRESS: Number of unpacked features : 9
PROGRESS: Number of coefficients    : 108
PROGRESS: Starting Newton Method
PROGRESS: --------------------------------------------------------
PROGRESS: +-----------+----------+--------------+-------------------+---------------------+
PROGRESS: | Iteration | Passes   | Elapsed Time | Training-accuracy | Validation-accuracy |
PROGRESS: +-----------+----------+--------------+-------------------+---------------------+
PROGRESS: | 1         | 2        | 4.801206     | 0.984853          | 0.984586            |
PROGRESS: | 2         | 3        | 7.523355     | 0.984853          | 0.984586            |
PROGRESS: | 3         | 4        | 10.246502    | 0.984853          | 0.984586            |
PROGRESS: | 4         | 5        | 12.992652    | 0.984853          | 0.984586            |
PROGRESS: | 5         | 6        | 15.856804    | 0.984853          | 0.984586            |
PROGRESS: | 6         | 7        | 19.004959    | 0.984853          | 0.984586            |
PROGRESS: | 7         | 8        | 21.787109    | 0.984853          | 0.984586            |
PROGRESS: | 8         | 9        | 24.578260    | 0.984853          | 0.984586            |
PROGRESS: | 9         | 10       | 27.447411    | 0.984853          | 0.984586            |
PROGRESS: | 10        | 11       | 30.333551    | 0.984853          | 0.984586            |
PROGRESS: +-----------+----------+--------------+-------------------+---------------------+

Model:
Class                         : LogisticClassifier

Schema
------
Number of coefficients        : 108
Number of examples            : 2577279
Number of classes             : 3
Number of feature columns     : 9
Number of unpacked features   : 9

Hyperparameters
---------------
L1 penalty                    : 0.0
L2 penalty                    : 0.01

Training Summary
----------------
Solver                        : auto
Solver iterations             : 10
Solver status                 : TERMINATED: Iteration limit reached.
Training time (sec)           : 31.2156

Settings
--------
Log-likelihood                : 193729.689

Highest Positive Coefficients
-----------------------------
OS[0]                         : 17.3944
OS[0]                         : 13.9204
OS[Windows]                   : 3.9543
TrafficType[site]             : 2.6564
CreativeType                  : 2.345

Lowest Negative Coefficients
----------------------------
(intercept)                   : -16.6472
AppSiteCategory[Arts & Entertainment#Society#Style & Fashion]: -8.4825
AppSiteCategory[Dating#Society]: -8.2136
AppSiteCategory[Health & Fitness]: -8.0631
AppSiteCategory[Education]    : -7.9956
	
	In [22]:

model.evaluate(test_data)

Out[22]:

{'accuracy': 0.9845860334946152, 'confusion_matrix': Columns:
 	target_label	str
 	predicted_label	str
 	count	int
 
 Rows: 3
 
 Data:
 +--------------+-----------------+--------+
 | target_label | predicted_label | count  |
 +--------------+-----------------+--------+
 |      c       |        0        |   16   |
 |      w       |        0        |  7926  |
 |      0       |        0        | 507305 |
 +--------------+-----------------+--------+
 [3 rows x 3 columns]}
	
	
	Feature engineering:
	

	
'TrafficType','AppSiteCategory','Position','Age','Gender','OS','DeviceType','Country','CreativeType'
	
model2=graphlab.logistic_classifier.create(products2,target='Outcome',features=['TrafficType','AppSiteCategory','Position','Age','Gender','OS','DeviceType','Country','CreativeType'],validation_set=test_data)
	
	
	
	
	usfull11=usfull[usfull['Age']!=0]
	usfull2=usfull1[usfull1['Gender']!='0']
	
	
	usfull2['predicted'] = model.predict(usfull2)
	
	===========================================================
	model1: 108 vars...  limited dataset 2.5gb, extend data
	model 1:  optimise models...
	
	
	=====================================================
	
	
	
	
	BidId,TrafficType,PublisherId,AppSiteId,AppSiteCategory,Position,BidFloor,Timestamp,Age,Gender,OS,OSVersion,Model,Manufacturer,Carrier,DeviceType,DeviceId,DeviceIP,Country,Latitude,Longitude,Zipcode,GeoType,CampaignId,CreativeId,CreativeType,CreativeCategory,ExchangeBid,OutcomeBidId,TrafficType,PublisherId,AppSiteId,AppSiteCategory,Position,BidFloor,Timestamp,Age,Gender,OS,OSVersion,Model,Manufacturer,Carrier,DeviceType,DeviceId,DeviceIP,Country,Latitude,Longitude,Zipcode,GeoType,CampaignId,CreativeId,CreativeType,CreativeCategory,ExchangeBid,Outcome
	
	
	
	
	
	A popular approach towards solving class imbalance problems is to bias the classifier so that it pays more attention to the positive instances. This can be done, for instance, by increasing the penalty associated with misclassifying the positive class relative to the negative class. Another approach is to preprocess the data by oversampling the majority class or undersampling the minority class in order to create a balanced dataset.