Data Schema

Data Schema are important to train an agent. A data schema defines which column of a dataset should be predicted, which ones are indepentent variable and which are switches that should be varied in order to minimze the cost function.

The data schema model

The agent model contains all the information about the agent.

Properties

Name
SwitchColumns
Type
json
Description
The columns which switch states should be optimzed during inference.
Name
IndependentColumns
Type
json
Description
The columns which might have an influence on the dependent variable.
Name
DependentColumns
Type
json
Description
The columns which should be predicted.
Name
HelperColumns
Type
datetime
Description
The columns which should not used during training, but might be used during inference.

Example Data Schema

{
  "SwitchColumns":       [
                              {"name": "hasAgofCampaign", "type":"binary", "options": {}, "switch_states": [0, 1]},
                              {"name": "hasIPTargetingCampaign", "type":"binary", "options": {}, "switch_states": [0, 1]},
                              {"name": "fcCappingCampaign", "type":"numerical", "options": {"log_transform": False, "min_value": [1], "max_value": [100], "epsilon": None}, "switch_states": [1, 5, 10, 20, 50]},
                              {"name": "fcCappingTime.1", "type":"numerical", "options": {"log_transform": True, "min_value": [10], "max_value": [14], "epsilon": [1]}, "switch_states": [1, 1*60, 1*60*30, 1*60*60, 4*60*60, 8*60*60, 24*60*60, 2*24*60*60,  8*24*60*60,  14*24*60*60]},
                              {"name": "maxCpm", "type": "numerical", "options": {"log_transform": False, "min_value": [0], "max_value": [1200], "epsilon": None}, "switch_states": [175, 1190,  350,  275,  276,  770,  330,  300, 1200,  340,  400, 375,  208,  250,  210, 500]},
                              {"name": "biddingStrategy", "type": "categorial", "options": {"categories": ["INTERVAL_BUDGET_OPTIMIZED", "MAXPRICE", "FLOORPRICE"]}, "switch_states": ["INTERVAL_BUDGET_OPTIMIZED", "MAXPRICE"]},
                              {"name": "deliveryTechnique", "type": "categorial", "options": {"categories": ["SMOOTH", "SIMPLE"]}, "switch_states": ["SMOOTH", "SIMPLE"]},
                              {"name": "hashAgofListCampaign", "type": "categorial", "options": {"categories": ["78dba1dda34d66a63363f3071c3bc542ec563f49ba72d1055f7f631f7267c8f0", "e448808417a4f48e3db3d77a8a316fd9f4c186976756f928cf9f92bc9e89cb3b", "2f363fddccdfbea43dc2760033caa4c1890a1f6cd8c7914abff51ec0bca5be44", "bd0b550608481e7e2cf7bb790589bff15543e5728bc067cb378968e407962ab2", "bab53f6a9302a236646f22c09584d543be48b828537251829e8dbbd14875d590", "1b1f52b15a73dcc3dd91f9512d623a8d740d53a994813dc74c4ce1b8157c90c6"]}, "switch_states": ["78dba1dda34d66a63363f3071c3bc542ec563f49ba72d1055f7f631f7267c8f0", "1b1f52b15a73dcc3dd91f9512d623a8d740d53a994813dc74c4ce1b8157c90c6"]},
                          ],
  "IndependentColumns":  [
                              {"name": "size", "type": "categorial", "options": {"categories": ['90x728', '600x160', '250x300', '250x970', '50x300', '75x300','50x20']}},
                              {"name": "hour", "type": "categorial", "options": {"categories": [ 5,  8, 11, 14, 17, 20, 23,  2]}},
                              {"name": "dayofweek", "type": "categorial",  "options": {"categories": [6, 0, 1, 2, 3, 4, 5]}},
                              {"name": "month", "type": "categorial", "options": {"categories": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]}},
                              {"name": "totalBudget", "type": "numerical", "options": {"log_transform": True, "min_value": [6], "max_value": [14], "epsilon": [1]}},
                              {"name": "totalImpressions", "type": "numerical", "options": {"log_transform": True, "min_value": [7], "max_value": [14], "epsilon": [1]}},
                          ],
  "DependentColumns":     [
                              {"name": "Clicks", "type": "numerical", "options": {"log_transform": True, "min_value": [0], "max_value": [5], "epsilon": [1]}},
                              {"name": "Impressions", "type": "numerical","options": {"log_transform": True, "min_value": [0], "max_value": [12], "epsilon": [1]}},
                              {"name": "Costs", "type": "numerical", "options": {"log_transform": True, "min_value": [-12], "max_value": [3], "epsilon": [0.00001]}},
                          ],
  "HelperColumns":        [
                              {"name": "targetImpressions", "type": "numerical", "options": {"scale": False, "log_transform": False, "min_value": [0], "max_value": [12], "epsilon": [1]}},
                          ]
  }

The column model

The column class represent a column in a dataset. The class covers information about data type and normalization, as well as switch states.

Properties

Name
name
Type
json
Description
The name of the column.
Name
type
Type
json
Description
The data type of the column, either binary, categorical or numerical.
Name
options
Type
json
Description
Normalization data for numerical and categorical columns (see example above).

Optional Properties

Name
switch_states
Type
array
Description
The state which are allowed the varry during inference.

Example Column

{
  "name": "size", 
  "type": "categorial", 
  "options": {
    "categories": ['90x728', '600x160', '250x300', '250x970', '50x300', '75x300','50x20']
    }
}