An Oracle White Paper

October 2013

Oracle Data Miner (Extension of SQL Developer 4.0)

Generate a PL/SQL script for workflow deployment

Denny Wong
Oracle Data Mining Technologies
10 Van de Graff Drive
Burlington, MA 01803
USA
[email protected]

Contents
Introduction .................................................................................................................................................. 3
Importing the Workflow ............................................................................................................................... 3
Create the User ......................................................................................................................................... 3
Load the Table........................................................................................................................................... 3
Import the Workflow ................................................................................................................................ 4
Workflow Overview ...................................................................................................................................... 5
Modeling ................................................................................................................................................... 5
Scoring....................................................................................................................................................... 6
Deployment Use Case ............................................................................................................................... 6
Workflow Run ............................................................................................................................................... 6
Generating Workflow Script Files ................................................................................................................. 7
Deploy Script Options ............................................................................................................................... 7
Generate Script UI Wizard ........................................................................................................................ 8
Script Files Specifications ........................................................................................................................ 11
Master Script: <Workflow name>_Run.sql ......................................................................................... 11
Cleanup script: <Workflow name>_Drop.sql ...................................................................................... 11
Workflow Image: <Workflow name>.png ........................................................................................... 11
Node Script: <Node name>.sql ........................................................................................................... 11
Running Workflow Script Files .................................................................................................................... 12
Variable Definitions................................................................................................................................. 12
Control Table........................................................................................................................................... 13
Table Structure.................................................................................................................................... 13
Column Descriptions ........................................................................................................................... 14
Scheduling Workflow Script Files ................................................................................................................ 17
SQL Developer......................................................................................................................................... 17
Oracle Enterprise Manager Jobs ............................................................................................................. 24
Conclusions ................................................................................................................................................. 29
Appendix ..................................................................................................................................................... 30

Introduction
Integrating data mining with an end user application has been one of the more challenging assignments
for developers of advanced analytics. With Data Miner 4.0, this effort has been made considerably
easier to accomplish. Data analysts and application developers can now work together to develop
advanced analytics and easily imbed them into applications.
Data analysts can use the Data Miner workflows to define, build and test their analytic methodologies.
When they are satisfied with the results, they can use the new script generation feature to hand off a
set of SQL scripts to the application developer. The application developer will be able to take these
standard Oracle SQL scripts and integrate them easily into their applications. Since all of the results of
Oracle Data Miner are database objects, there is no concern about moving data and models from non
Oracle database systems.
Although the primary audience for this paper are application developers; data analysts, database
administrators and IT management can benefit from reading it as well.
Objectives/Benefits:

Application Developers
o Describes the structure of the generated scripts and run time behavior.
o Shows how the generated script can be scheduled to run in the database.
Data Analysts
o Describes how to generate a SQL script for all or part of a workflow.
Database Administrators and IT Management
o Provides an understanding of the entire process and how it all runs within the Oracle
Database environment.

The business benefit will be in the ability to quickly integrate advanced analytics into existing
applications without incurring the high cost and complexity of integrating additional non-database
platforms

Importing the Workflow

Create the User
This white paper demo requires a Data Miner installation and a Data Miner user account. Please refer
to the Oracle By Example Tutorials to review how to create a Data Miner user account and install the
Data Miner Repository.

Load the Table

The example in this white paper involves a demo database table INSUR_CUST_LTV_SAMPLE, which
comes with the Data Miner Repository installation by default. If it is not loaded in your user account,
you can install it manually using the instInsurCustData.sql SQL script (available in the companion .zip
download for this white paper).
3

The instInsurCustData.sql SQL script can be invoked in the SQL Worksheet of SQL Developer:

Make sure that the script is run from within the user schema.

Import the Workflow

The Oracle Data Miner interface provides a mechanism for exporting and importing workflows that
capture complex data mining methodologies. The next step in this example is to create a project and
import a pre-defined workflow into existing Data Miner user account. The workflow,
codegen_workflow.xml (available in the companion .zip download for this white paper) will be imported
into the user account as shown below.
To create the project and import the workflow, right-click in the Data Miner navigator as shown in the
two images below. The first step will create a new project, and the second will perform the import of
the pre-defined workflow, codegen_workflow.xml.

Once imported, the workflow should look like the picture below:

Workflow Overview
The workflow is comprised of two distinct processes contained within a single lineage: modeling (top)
and scoring (bottom). Both processes use the demo data INSUR_CUST_LTV_SAMPLE as input data
source.

Modeling
The modeling process is used for building a classification SVM model to predict whether the customer
will buy insurance or not. The model coefficients are persisted to a database table for viewing and this
table may provide a basis for application integration.

Scoring
The scoring process is used to make predictions for the customer data using the SVM model created by
the modeling lineage. The prediction result is persisted to a database view for viewing. This view
always reflects the predictions of the current input data. For example, if the input table is refreshed
with new data, this view will automatically capture the predictions of the new data. Moreover, this view
may provide a basis for application integration.

Deployment Use Case

For this demo, we assume a Data Analyst defines the above workflow for model build and score
purposes. When he is satisfied with the results, he then uses the new script generation feature to hand
off a set of SQL scripts to the Application Developer for deployment. The Developer deploys the scripts
to the target or production database, where he can schedule the scripts to run periodically, so that the
model will be rebuilt with fresh customer data. In this demo, you will see how to use both SQL
Developer/Oracle Scheduler and Oracle Enterprise Manager to schedule the scripts to run on the
database. Although not described in this paper, the output results generated by the scripts can easily be
integrated into end user reporting applications like Oracle Business Intelligence Enterprise Edition. The
output results are described below to give you some ideas on what insights they could provide for the
end user.

MODEL_COEFFICIENTS underlying database table can be queried to provide insight on what

predictors are used to determine whether the user will buy insurance or not.
SCORED_CUSTOMERS underlying database view can be queried to display whether a customer
will purchase insurance or not, and the likelihood of the prediction. This view always reflects
the real time prediction of current customer data.

Workflow Run
The workflow must be run before it can be deployed. Right-click the INSUR_CUST_LTV_SAMPLE BUILD
node and select Force Run | Selected Node and Children from the menu to run the whole workflow.

Once workflow run completed, the workflow should look like the picture below:

Generating Workflow Script Files

The next step in this demo is to generate a SQL script from the workflow.

Deploy Script Options

There are three deploy options of how the script can be generated:

Selected node and dependent nodes

o Generate script for the selected node(s) and all its parent nodes
o For example, if the Apply node is selected, a script will be generated for these nodes:
INSUR_CUST_LTV_SAMPLE BUILD, Class Build, INSUR_CUST_LTV_SAMPLE APPLY, and
Apply.
Selected node, dependent nodes, and children nodes
o Generate script for the selected node(s) and all its parent and children nodes

For example, if the Apply node is selected, a script will be generated for these nodes:
INSUR_CUST_LTV_SAMPLE BUILD, Class Build, INSUR_CUST_LTV_SAMPLE APPLY, Apply,
and SCORED_CUSTOMERS.
Selected node and connected nodes
o Generate script for the selected node(s) and all nodes that are connected to this node.
o For example, if the Apply node is selected, a script will be generated for all nodes in the
workflow.

Alternatively, to generate a script for the entire workflow, you can multi-select all nodes in the workflow
and select any of above options from the selected node.
To generate a script for the demo workflow, right-click the INSUR_CUST_LTV_SAMPLE BUILD node and
select the Deploy | Selected node and connected nodes item from the menu.

Generate Script UI Wizard

After you select the deploy option, the Generate SQL Script wizard launches. You must select Target
Database Version, so that the generated script will be compatible for that database version. If the
workflow nodes included in the script operation contain features that are not compatible with the target
database, a warning dialog will be displayed along with unsupported feature information.

Next you need to specify the Script Directory where the scripts will be saved. The Script Directory name
defaults to the workflow name. A new directory will be created to store all the generated scripts.

Click Finish to generate the script.

10

Script Files Specifications

The generated script consists of the following script files:
Master Script: <Workflow name>_Run.sql
The Master Script invokes all the required node level scripts in the appropriate order. It performs
some validation, such as testing to determine if the version of the script is compatible with the
version of the Data Miner Repository that is installed. It creates a workflow master table that
contains entries for all the underlying objects created by the workflow script.
Cleanup script: <Workflow name>_Drop.sql
The Cleanup Script drops all objects created by running the master script. It drops hidden objects
(such as the table generated for Explore Data) as well as public objects (such as Model Names
created by Build nodes, and Tables created by a Create Table node).
Workflow Image: <Workflow name>.png
An image (.png file) of the workflow at the time the script was generated. The entire workflow is
displayed.
Node Script: <Node name>.sql
A Node Script performs node specific operation (such as model creation in Build node). A node
script is generated for each node that participates in the script generation. The node script should
be invoked by the master script only; it should not be invoked directly. Refer to the Appendix for
output result for each node type.
For the demo workflow, the following script files are generated:

Master script: codegen_workflow_Run.sql.

Cleanup script: codegen_workflow_Drop.sql.
Workflow image: codegen_workflow.png.
Node scripts: Apply.sql, Class Build.sql, Coefficients.sql, INSUR_CUST_LTV_SAMPLE APPLY.sql,
INSUR_CUST_LTV_SAMPLE BUILD.sql, and MODEL_COEFFCIENTS.sql, and SCORED_CUSTOMERS.sql.

11

Running Workflow Script Files

Variable Definitions
The Node level scripts have variable definitions that provide object names for the public objects created
by the scripts. The Master script is responsible for invoking all the underlying node level scripts in proper
order, so all variable definitions must be defined in the Master script.
The following variables are supported:

Variables that allows you to change the name of the object names that are input to the Node
level scripts, such as tables/views and models. By default, these names are the original
table/view and model names.
Variable that allows you to change the name of the Control table (see below). By default, this
name is the workflow name.
Variable that indicates if named objects should be deleted first before they are generated by the
script.

The following object types can be named using these definitions:

Models referenced in Build Nodes or in a Model Node

Tables referenced in Create Table or Update Table Nodes

For the demo workflow, the following variable definitions are generated in the Master script:
-- Subsitution Variable Definition Section: Override default object names here
DEFINE WORKFLOW_OUTPUT = 'codegen_workflow'
-- Drop user named objects (e.g. model, output table)? TRUE or FALSE
DEFINE DROP_EXISTING_OBJECTS = 'TRUE'
-- From Node: "INSUR_CUST_LTV_SAMPLE BUILD"
DEFINE DATA_SOURCE_1 = '"DMUSER"."INSUR_CUST_LTV_SAMPLE"'
-- From Node: "INSUR_CUST_LTV_SAMPLE APPLY"
DEFINE DATA_SOURCE_2 = '"DMUSER"."INSUR_CUST_LTV_SAMPLE"'
-- From Node: "Class Build"
DEFINE MODEL_1 = '"DMUSER"."CLAS_SVM_MODEL"'
-- From Node: "MODEL_COEFFCIENTS"
DEFINE CREATE_TABLE_2 = '"DMUSER"."MODEL_COEFFCIENTS"'
-- From Node: "SCORED_CUSTOMERS"
DEFINE CREATE_VIEW_3 = '"DMUSER"."SCORED_CUSTOMERS_V"'
The Control table name variable, WORKFLOW_OUTPUT defaults to the workflow name. The drop
variable, DROP_EXISTING_OBJECTS defaults to TRUE, which indicates all existing named objects should
be removed before they are generated. The Node specific object variables default to their original
names in the workflow. For example, the generated Node variable DATA_SOURCE_1 allows user to
override the input table used in the INSUR_CUST_LTV_SAMPLE BUILD Data Source node. It is expected

12

that the input data sources have the same columns as the originally referenced input data sources.
Missing column names could result in a run time failure.

Control Table
When the Master script is run, it first creates the Control table using the name specified in the control
table name variable. The purposes of the Control table are the followings:

Generated objects, such as views, models, text specifications, are registered in this table.
Logical nodes in the workflow are able to look up their input objects and register their output
objects in the control table.
The Cleanup script uses the table to determine what objects need to be dropped.
For advanced users, the Control table provides internal name of objects that are not readily
accessible via the workflows today. For example, users can find the model test result tables by
viewing the Control table.
By using different control file names along with different output variable names, the generated
script can be used concurrently to generate and manage different results. This may be useful if
the input data sources continue different sets of data that you wish to mine independently. In
this use case, the application would be responsible for saving the name of the control table so
that it can be utilized when rerunning or dropping the generated results.

Table Structure
The Control table is defined as following structure:
CREATE TABLE "&WORKFLOW_OUTPUT"
(
NODE_ID VARCHAR2(30) NOT NULL,
NODE_NAME VARCHAR2(30) NOT NULL,
NODE_TYPE VARCHAR2(30) NOT NULL,
MODEL_ID VARCHAR2(30),
MODEL_NAME VARCHAR2(65),
MODEL_TYPE VARCHAR2(35),
OUTPUT_NAME VARCHAR2(30) NOT NULL,
OUTPUT_TYPE VARCHAR2(30) NOT NULL,
ADDITIONAL_INFO VARCHAR2(65),
CREATION_TIME TIMESTAMP(6) NOT NULL,
COMMENTS VARCHAR2(4000 CHAR)
)

13

Column Descriptions
The followings describe how the columns in the table are used:
Column Name

Description

Examples

NODE_ID

Workflow Node id. This uniquely

identifies each node.

10001, 10002

NODE_NAME

Workflow Node name.

Class Build, MINING_DATA_BUILD_V

NODE_TYPE

Node type is the category of node

type that the node is based on.

Data Source, Class Build

MODEL_ID

Workflow Model id. This uniquely

identifies each model referenced
within a workflow.

10101, 10102

MODEL_NAME

Name of Model.

CLAS_GLM_1_6

MODEL_TYPE

Model type is the algorithm type

used by the model.

Generalized Linear Model

OUTPUT_NAME

Table/View Name, Model Name, Text object

names
The name of the output. These will
Such as:
be internally generated names
ODMR$15_37_21_839599RMAFRXI - table
unless the names are under the
name
control of the user.
"DMUSER"."CLAS_GLM_1_6" - fully qualified
model name

OUTPUT_TYPE

Qualifies what type of object.

ADDITIONAL_INFO

Used to qualify what the purpose of

the object is in context of the script Target class for test lift result
execution.

CREATION_TIME

Object creation time.

11-DEC-12 03.37.25.935193000 PM (format

determined by locale)

Comment to qualify the object's

role in the context of the script
execution.

Output Data (displayed for nodes like Data

Source)
Data Usage (displayed for the view passed
into model build)
Weights Setting (displayed for a weights
table passed into model build)
Build Setting (displayed for a build settings
table passed into model build)
Model (displayed for a Model object)

COMMENTS

14

Table, View, Model

To run the deployed workflow, invoke the Master script in the SQL Plus:
>@" C:\code gen\codegen workflow\codegen_workflow_Run.sql"
For subsequent run, invoke the Cleanup script first to delete previously generated objects, and then run
the Master script:
>@" C:\code gen\codegen workflow\codegen_workflow_Drop.sql"
>@" C:\code gen\codegen workflow\codegen_workflow_Run.sql"
After the script is run successfully, you can query the Control table to examine the generated objects:
>select * from "codegen_workflow"

For example, the Create Table Node, MODEL_COEFFICIENTS, produced an output table
MODEL_COEFFCIENTS that persisted the coefficient data extracted from the generated SVM model.

15

To examine the coefficient data, you can query the output table:
> select * from "DMUSER"."MODEL_COEFFCIENTS"

16

Scheduling Workflow Script Files

We will use Oracle Database to schedule the SQL script files to run on the target or production database.
All generated SQL script files need to be deployed to the target or production database host machine
where they are accessible by the database instance, and these files should be stored together in the
same directory. In addition, Data Miner repository is required to run the script files because some node
scripts make use of repository provided services at runtime. Moreover, Data Miner user account is
required to run the script files because it has necessary grants to the repository provided services. In
the following sections, we will show you how to use both SQL Developer and Oracle Enterprise Manager
web application to schedule the master script to run.
Add full directory path in the master file
A full directory path needs to be added to each node script invocation in the master script, so that these
individual node script files can be called by the master script during runtime.
In the master script, codegen_workflow_Run.sql, find the following node script invocations:
-- Workflow run
@@"INSUR_CUST_LTV_SAMPLE BUILD.sql";
@@"INSUR_CUST_LTV_SAMPLE APPLY.sql";
@@"Class Build.sql";
@@"Coefficients.sql";
@@"MODEL_COEFFCIENTS.sql";
@@"Apply.sql";
@@"SCORED_CUSTOMERS.sql";
Then add the full directory path where the scripts will be stored in the target or production database
host machine. In this example, we assume the script files will be deployed to the /home/workspace
directory.
-- Workflow run
@@"/home/workspace/INSUR_CUST_LTV_SAMPLE BUILD.sql";
@@"/home/workspace/INSUR_CUST_LTV_SAMPLE APPLY.sql";
@@"/home/workspace/Class Build.sql";
@@"/home/workspace/Coefficients.sql";
@@"/home/workspace/MODEL_COEFFCIENTS.sql";
@@"/home/workspace/Apply.sql";
@@"/home/workspace/SCORED_CUSTOMERS.sql";

SQL Developer
SQL Developer provides a graphical interface for developers to define Scheduler Jobs. A SQLPlus script
job is required to run the SQL script files, but the Script job type is only supported in the 12c database.
The Job definition defines the master script invocation as a script file using a full file path. The user can

17

decide on whether the Job should be run on a schedule or on demand. The Job execution can be
monitored within the application. The result will either be a success or a reported failure.
The following system privileges are required for the account that runs the scheduled SQL script files:

CREATE CREDENTIAL
CREATE EXTERNAL JOB

A credential is an Oracle Scheduler object that has a user name and password pair stored in a dedicated
database object. A SQLPlus script job uses a host credential to authenticate itself with a database
instance or the operating system so that the SQLPlus executable can run. In addition, the job may point
to a connect credential that contains a database credential, which is used to connect SQLPlus to the
database before running the script.
To create credentials, right-click the Credentials item in the Connections navigator as shown below.

18

First we will create the database host credential, and then the database credential.

Next we will use the Job wizard to define a new Scheduler job. Click the Job item in the Connections
navigator to launch the wizard.

19

In the Job Details step, enter the following Job name and description. Select the Script job type and
SQLPlus script type, and enter the full path names of the cleanup and master scripts in the script
window. You can specify if the job will be run one time only or repeating. The following shows the job
will be scheduled to run on the first Sunday of each month at midnight.

20

In the Destination step, select the Local database and the credentials created above.

21

In the Notification step, you can set up an email notification based on the job status. For this to work,
email server should be set up in the host machine. For now, we will skip to the Summary step and click
Finish to create the job.

Once the job is created, you can monitor it within the SQL Developer. To monitor the job, right-click the
job name, CODEGEN WORKFLOW, in the Jobs folder in the Connections navigator to open the job viewer
as shown below.

22

Click the Run Log tab to see the status of the job. Here you can see when the job started running, how
long it took to complete, CPU usage, etc. In case of failure, you should see an error message.

To examine the script run result, click the Job Script Output tab in the bottom window, select the result
in the OUTPUT column and right-click to bring up the context menu. Then select the Single Row View to
bring up the viewer with the script run result.

23

Oracle Enterprise Manager Jobs

Oracle Enterprise Manager allows DBAs to define Jobs through the application. The Job definition
defines the master script invocation as a script file using a full file path. The user can decide on whether
the Job should be run on a schedule or on demand. The job execution can be monitored in the
application. The result will either be a success or a reported failure.
User needs to log in to the Oracle Enterprise Manager using the DBA account.

24

Once logged in, use the Jobs link in the Job Activity section at the bottom of the page to launch the Job
creation page.

Select the SQL Script type in the Create Job drop-down list and click Go. This will take you to the Create
Job page where you can define the new job.

25

In the General tab, enter the following Job name and description and add a target database where the
job will be run.

In the Parameters tab, enter the full path names of the cleanup and master scripts in the SQL Script edit
control.

26

In the Credentials tab, enter the database host credential and the database credential.

In the Schedule tab, you can specify if the job will be run one time only or repeating. The following
shows the job will be scheduled to run on the last day of each month at midnight.

27

In the Access tab, you can set up an email notification based on the job status. For this to work, email
server should be set up in the host machine.

Once you have entered all the settings, click Submit to create the job.

28

Conclusions
This white paper shows how easy it is to deploy Data Miner workflow to the target or production
database using the SQL script generation feature. Furthermore, Oracle Enterprise Manager in the target
database can be used to schedule the scripts to run in any desirable time interval. Given that Oracle's
data mining and advanced analytics operate natively inside the Oracle Database, mining insights and
predictions also remain inside the database so can be accessed by SQL queries from OBIEE. Predictive
model results can be called interactively using OBIEE reports and dashboards. With Oracle, all the data
mining and advanced analysis and results stay inside the database providing a simpler architecture,
better security, better scalability and a single source of truth.

29

Appendix
The following table describes the objects that are generated by each node script.
Script Type
Create Table Node
Data Explore Node
Aggregate Node
Anomaly Detection Query Node (12c)
Apply Node
Apply Text Node
Clustering Query Node (12c)
Data Source Node
Feature Extraction Query Node (12c)
Filter Columns Details Node
Filter Rows Node
Join Node
Model Details Node
Prediction Query Node (12c)
Sample Node
SQL Query Node
Transform Node
Filter Columns Node

Output
Creates a table or view that persists the input data
Creates a table that contains statistical data of the
input data
Creates a view reflecting the output of the node

Build Text Node

Classification Build Node

Creates a view reflecting the output of the Column

Filter node (like other transform type nodes)
If the Attribute Importance setting is specified, a
table is generated containing the AI result
For each text transformation the following objects
are created
o Feature Table
o Oracle Text Objects (Policy, Lexer, Stoplist)
Creates a view reflecting the output of the Build
Text node. This is essentially the same output as
the Apply Text node
Create a Build and Test (if necessary) data
For each imbedded text transformation (12c),
Oracle Text Objects are created (Policy, Lexer,
Stoplist)
A model is created for each model build
specification
GLM Model Row Diagnostics Table (if row
diagnostics turned on)
Each Model Test will have one table generated for
each of the following test results
o Test Metric
o Confusion Matrix
o ROC (binary classification only)
30

Regression Build Node

Anomaly Detection Build

Clustering Build
Feature Extraction Build

Association Build

Test Node (Classification)

Test Node (Regression)

Model Node
Text Reference Node
Graph Node

Each Model Test will have one table for each of the
following tests per target value (up to 100
maximum target values)
o Lift
Create a Build and Test (if necessary) data
For each imbedded text transformation (12c),
Oracle Text Objects are created (Policy, Lexer,
Stoplist)
A model is created for each model build
specification
GLM Model Row Diagnostics Table (if row
diagnostics turned on)
Each Model Test will have one table generated for
each of the following test results
o Test Metric
o Residual Plot data
For each imbedded text transformation (12c),
Oracle Text Objects are created (Policy, Lexer,
Stoplist)
A model is created for each model build
specification
A model is created for each model build
specification
Same specifications as Class Build except no models
are created
Same specifications as Regression Build except no
models are created
No code generated

31