A data mining analyst wants to obtain the rules contained
in historic data. A specific mining algorithm is selected, configured,
and applied to a specified set of input data. The execution of the
mining algorithm is called training phase. The result of the training
phase is a data mining model.
You can use the model for the following tasks:
- Computing scores for new records (applying the model)
- Graphically exploring the model with the Intelligent Miner® Visualizer)
- Extracting the rules as database rows by using Intelligent Miner table functions
In Figure 1 the Modeling buildModel function receives a BldTask object that describes the physical (MiningData) and logical (LogicalDataSpec) schema of the input table as well as the settings that configure the mining algorithm. It produces a mining model.
Figure 1. SQL API in Intelligent
Miner
The training phase consists of the following steps:
- Specifying mining data
Consists of source table name, column names, and types, optional alias names
- Defining logical data specifications
Holds for each field in the source table the mining field name and mining field type. It also optionally holds name mappings, taxonomies, or numerical field properties. A mining field name matches an alias name in mining data to map the fields during the training run.
- Defining mining settings Consists of the logical data specification and additional mining function/algorithm, specific settings, such as:
- Transaction/item field and confidence and support for Association
- Target field for Classification or Regression
- Number of clusters for Clustering
- Defining mining tasks Combines mining settings with mining data. Additionally some control parameters can be set such as the location for error or progress information. For example:
INSERT INTO IDMMX.ClasTasks (ID,TASK) WITH MyData (MyDataCol) AS (VALUES( IDMMX.DM_MiningData()..DM_defMiningData('HEART'))) SELECT 'HeartTask', IDMMX.DM_ClasBldTask().. DM_defClasBldTask(MyDataCol,cast(NULL as IDMMX.DM_MiningData), IDMMX.DM_ClasSettings()..DM_useClasDataSpec(MyDataCol..DM_genDataSpec()).. DM_setClasTarget('DISEASED')) FROM MyData; - Building and storing mining models Executing a predefined mining task. The mining model is written into a table during execution. For example:
CALL IDMMX.DM_buildClasModelCmd( 'IDMMX.CLASTASKS', 'TASK','ID','HeartTask', 'IDMMX.CLASSIFMODELS','MODEL','MODELNAME','CustModel');