You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A lot of databases have non verbal or non descriptive column names that will for sure confuse semantically the LLM.
Would it be possible to add a method to describe or alias the column of a table ?
For example let's say I have a table called bsd with a column called cst: I want to alias them or describe them as building system and customer transaction.
Hopefully makes sense.
The text was updated successfully, but these errors were encountered:
For our applications, we have chosen to define what these column names mean, when appropriate, as documentations, and train vanna to know what these column names are and what they mean. I would think such an approach would work for you as well.
Just remember to store each column name document chunk in a format that makes sense and preserves the hierarchical nature of your database, for instance like so: {schema, table, columnName, description}.
It is also likely that you might want to store chunks on table-level instead of row-level. Depends on your database and use case, I would think, what makes the most sense.
vn.train(documentation="the table=Identity has column=id, the object unique identifier")
vn.train(documentation="the table=Identity has column=created, the date when the object was created")
....
A lot of databases have non verbal or non descriptive column names that will for sure confuse semantically the LLM.
Would it be possible to add a method to describe or alias the column of a table ?
For example let's say I have a table called bsd with a column called cst: I want to alias them or describe them as building system and customer transaction.
Hopefully makes sense.
The text was updated successfully, but these errors were encountered: