Oracle E-Business Suite and 23ai Vectors: Examples

Oracle 23ai will profoundly impact the Oracle E-Business Suite. Specifically, Vectors, a new datatype introduced by 23ai, will change how questions can be asked and answered. While many have already posted about 23ai, I am sharing what I think are a few of the mechanical steps that will need to transpire within an EBS database to create, populate, and then use Vectors.

Creating Vectors

Vectors are new a datatype within the Oracle database. Like we have had integer and Varchar2 datatypes, Vectors are a new datatype. New columns will be added to tables of type Vector. For example, EBS inventory parts and their descriptions, if I wanted to add a Vector to the long description, I could use the following:

ALTER TABLE INV.MTL_SYSTEM_ITEMS_TL ADD ( item_desc_vector VECTOR, ?item_desc_long_vector VECTOR);

The question is, though, who will create the Vectors? With the EBS database comprising 25,000+ tables, my guess is that the same team that has been creating the EBS tables and Indexes for the past 30 years will be creating Vectors—the EBS product development team. My guess is that they are hard at work as we speak (or so I hope) creating these Vectors.

Populating Vectors

As with any new column added to a table, the Vectors columns will be blank until populated. How do you populate Vectors columns using existing data stored in EBS tables? Also, once populated, Indexes on the newly created populated Vector columns will be required. Who will populate the Vectors and who will build the Indexes?

Again, my guess is that the Oracle EBS product development team is actively working on creating and populating Vectors as well as as building solutions for keeping the "embedded" vector data current.

The actual steps required to populate a Vector are something like this:

UPDATE ININV.MTL_SYSTEM_ITEMS_TL item_desc_long_vector = DBMS_VECTOR.vector_embedding( all_MiniLM_L12_v2 using LONG_DESCRIPTION) as data);

Using Vectors

Vectors will fundamentally change how EBS data will be used, but how? Somebody (most likely at first EBS Product development) will need to write SQL to utilize the new data type. Developers of customizations (e.g., myself and others) will then follow. With 25,000+ tables comprising the Oracle EBS database, there will be the potential to write many 10,000s of new SQL statements to make use of the 1,000s of new Vectors. This will take time. However, where will this new SQL be written? Existing EBS Forms and Reports? Maybe. What about Oracle Analytics Cloud (OAC)? I am keen to think that OAC will play an essential part in how EBS Vectors will be used and accessed.

Here is an example of an SQL select statement making use of a hypothetical new vector column on the EBS table PER_ALL_PEOPLE_F -

SELECT person_id, VECTOR_DISTANCE(person_vector, ‘a man named mike born in the 1960s’, 'COSINE') as distance FROM per_all_people_f ORDER BY distance ASC

Closing comments

I wanted to share a few thoughts about what I see will most likely be occurring with regard to Oracle 23ai and EBS. When will this new functionality be forthcoming? I have no information but hope soon, ideally within 2025. Once I learn more, I will share. If any of you hear something, please share it too.

Please contact me with any questions: [email protected]

References

Vector User Guide: https://docs.oracle.com/en/database/oracle/oracle-database/23/vecse/ai-vector-search-users-guide.pdf

SQL Reference Guide: https://docs.oracle.com/en/database/oracle/oracle-database/23/vecse/vector_embedding.html#GUID-5ED78260-6D21-4B6B-86E0-A1E70EFA11CA

Excellent YouTube: https://www.youtube.com/watch?v=ZLGNKjDKBWU&t=699s&ab_channel=FahdMirza

DBMS_VECTOR https://docs.oracle.com/en/database/oracle/oracle-database/23/arpls/dbms_vector1.html#GUID-8E615832-F6C0-4435-8F43-3FAF80692D5B

Quick start: https://docs.oracle.com/en/database/oracle/oracle-database/23/vecse/sql-quick-start-using-vector-embedding-model-uploaded-database.html

eTRM: https://etrm.live/etrm-12.2.2/etrm.oracle.com/pls/trm1222/etrm_pnav27af.html?


#EBS, #23ai

要查看或添加评论,请登录

Mike Miller的更多文章

社区洞察

其他会员也浏览了