2
$\begingroup$

I want to check an algorithm that learns the connections between different tables in a database.

Where can I find datasets that are compound of multiple tables (i.e. more than 5) that have connections between them?

All I can find on the web (mainly on Kaggle.com) are datasets with 1 or 2 tables.

I'm aware that my question is kind of off topic but still I'm sure it can help people that have the same problem as I do

$\endgroup$
1
$\begingroup$

There are many places:

this one is very good: https://relational.fit.cvut.cz/

there are also:

https://datahub.io/search

https://www.reddit.com/r/bigquery/wiki/datasets

$\endgroup$
0
$\begingroup$

Is there any reason why you couldn't create them yourself? I assume your algorithm is doing things like looking for primary and foreign keys, so the actual data in the tables shouldn't really matter.

I would treat this as an exercise in data augmentation and just synthesize a series of tables with all kinds of keys. Some would have integer keys, others GUIDs, others some sort of hash column, etc, etc. It would be fairly straightforward to create a script that can create you any number of tables with matching keys like this.

$\endgroup$
1
  • $\begingroup$ The algorithm ignores key constraints and looks only on the data, I can synthesize data for a development environment, but real-world data is much better for the end to end tests. $\endgroup$ – yossico Jan 10 '19 at 9:18

Not the answer you're looking for? Browse other questions tagged or ask your own question.