With the release of the new Oracle 12c database and SQL Developer 4 we have a range of Oracle Data Miner new features . Some of these are embedded into the database and are only available in 12c. Check out my previous blog post on these new features.
In this blog post I will look at the new Oracle Data Miner features that come with the ODM tool in SQL Developer4.
The new features of the Oracle Data Miner tool can be grouped into 2 categories. The first category contains the new features that are available to all user of the tool (11.2g and 12c). The second category contains the new features that are only available in 12c. The new features of each of these categories will be explained below.
Category 1 – Common new features for 11.2g and 12c Database users
There is a new View Data feature that allows you to drill down to view the customer object and to view nested tables.
A new Graph Node that allows you to create graphs such as line, bar, scatter and boxplots for data at any stage of a workflow. You can specify any of the attributes from the data source for the graphs. You don’t seem to be limited to the number of graphs you can create.
A new SQL Node. This is welcome addition, as there has been many times that I’ve need to write some SQL or PL/SQL to do a specific piece of processing on the data that was not available with the other nodes. There are 2 important elements to this SQL node really. The first is that you can write SQL and PL/SQL code to do whatever processing you want to do. But you can only do it on the Data node you are connected to.
The second is that you can use it to call some ORE code. This allows you to use the power of R and extensive range of packages that are available to expand the analytic functionality that is available in the database. If there is some particular function that you cannot do in Oracle and it is available in R, you can now embed this function/code as an ORE object in the database. You can then called using SQL.
WARNING: this particular feature will only work if you have ORE installed on your 220.127.116.11g or 12.1c database
New Model Build Node features, include node level text specifications for text transformations, displays the heuristic rules responsible for excluding predictor columns and being able to control the amount of classification and regression test results that are generated. I’ll be covering these in later blog posts.
New Workflow SQL Script Deployment features. Up to now the workflow SQL script, I found to be of limited use. The development team have put a lot of work into generating a proper script that can be used by developers and DBA. But there are some limitations still. You can use the script will run the workflow automatically in the database without having the use the ODM tool. But it can only be run the in the schema that the workflow was generated. You will still have to do a lot of coding (although a lot less than you used to) to get your ODM models and workflows to run in another schema or database.
This will output the script to a file buried deep somewhere inside you SQL Developer directory. Unfortunately in the EA1 release, the size of this location field is small and scrolling has not been enabled. So you cannot (currently) scroll to the end of the field to see the actual location. You can edit this location to have a different shorter location.
Maybe this will be fixed for the official release.
Category 2 – New features for 12c Database users.
Now for the new features that are only visible when you are running ODM / SQL Dev 4 against a 12c database. No configuration changes are needed. The ODM tool checks to see what version of the database you are logging into. It will then present the available features based on the version of the database.
New Predictive Query nodes allows you to build a node based on the new non-transient feature in 12c called Predictive Queries (PQs). In SQL Developer we get 3 addition types of Predictive Queries. These can be used for Anomaly Detection, Clustering and Feature Extraction
It is important to remember that underlying model produced by these PQs to not exist in the database after the query has executed. The model is created, used on the data and then the model deleted.
The Clustering node has the new algorithm Expectation Maximization in addition to the existing algorithms of K-Means and O-Cluster.
The Feature Extraction node has the new algorithm called Principal Component Analysis in addition to the existing Non-Negative Matrix Factorization algorithm.
Text Transformations are now built into the model build nodes. These text transformations will be part of the Automatic Data Processing steps for the model build nodes. This is illustrated in the above images.
The Generalized Linear Model that is part of the Classification Node has a Feature Selection option in the Algorithm Settings. The default setting is Ridge Regression. Now there is an additional option of using Feature Selection.
Prediction Result Explanations gives the scoring details used to to explain why the prediction was made.
Look out for blog post on each of these new features.