IBM SQL Data Insights Quick Guide to Use Cases

Mapping IBM SQL Data Insights for Db2 v13 for z/OS to use cases linked to the scalar functions provided by Db2

After installing IBM SQL Data Insights (SDI from now on) coming as a feature with your Db2 v13 you will start thinking about possibilities to put it to work. Then you realise that there are so many options that you don’t know where to start. This is the place to understand the main use cases where you can use AI by just using your SQL skills.

By the end of this post, you will find some links to What is SQL Data Insights and why you should know it  and  SDI documentation that you might want to read. 

As the title states, this is a quick guide for you to start thinking about which is the data you need and how to exploit with SDI. These are general examples, a kind of inspiration for you to reflect on specific use cases useful for your company.

Use Cases

All cases have some characteristics in common that make them good for SDI:

• Data in table/view is mainly stable. Very dynamic tables will, inevitably, need frequent training for the models to update vector tables.
• All of them are intended for segmenting, comparing or grouping data. SDI does this very well.
• Data needed for the use cases should be (mainly) already in your Db2 for z/OS, so that you can easily create an ad-hoc table to train the models if no table already exist that contains what you need.
• The data types I put in the table are generic. You will need to dig into your data in Db2 to build the full use case.

When training

Some advice on managing columns and tables for training:

1. Columns with a greater number of unique values are the ones contributing most to add semantic meaning to the vectors.

If you need other columns in your table that have a small number of unique values compared to the amount of rows, you can un-select them before training since they will add no semantically significative meaning and you will save training time and CPU cycles. It might be the case that you need them in the result set to be able to fully understand the data returned, no problem, just don’t train with those columns. 

2. Having a primary key will help a lot to semantically distinguish rows and can make intelligent queries more effective. If your table does not have a primary key, generate a new column with it. Primary keys will reinforce vector discrimination during the training process and will help in the evaluation process when intelligent queries are used. 

3. Work out NULLs a little bit. Reduced training times and CPU cycles will be the recompense for that work. You have a place where to mark NULLS and values that you want to omit just before launching a training. It is in the UI of SDI. Telling the models what is a NULL in a column will add value to the vectors built and stored.

Links

In this Community:

What is SQL Data Insights and why you should know it

Documentation Links

• IBM SQL Data Insights Overview
• Installation and Configuration Guide
  

#Db2forz/OS #Db2Z #SDI