Add squared Euclidean distance (l2_squared) in java implementation

[zhichenxu-meta]

Description

This PR introduces the squared Euclidean distance (l2_squared) function between identical sized vectors represented as arrays(real). The l2_squared distance is commonly used to measure similarities between embeddings of multimedia data.

Differential Revision: D77157252

Motivation and Context

We are introducing vector search capabilities into Presto, and this PR takes the first step by adding common distance functions. This functionality will enable users to perform efficient similarity searches on multimedia data.

Impact

The addition of the l2_squared distance function will enhance Presto's capabilities in handling multimedia data and enable users to perform more complex analytics tasks.

Test Plan

• Unit tests
• Manual tests: on test servers with various input scenarios to ensure correctness and performance.

Contributor checklist

• Please make sure your submission complies with our contributing guide, in particular code style and commit standards.
• PR description addresses the issue accurately and concisely. If the change is non-trivial, a GitHub Issue is referenced.
• Documented new properties (with its default value), SQL syntax, functions, or other functionality.
• If release notes are required, they follow the release notes guidelines.
• Adequate tests were added if applicable.
• CI passed.

Release Notes

== RELEASE NOTES ==
General Changes

• Add l2_squared function to calculate the squared Euclidean distance between two identically sized vectors represented as arrays. This function supports both array(real) and array(double) input types. For more information, refer to the Euclidean distance definition.

Example Usage:

-- Using real arrays
SELECT l2_squared(ARRAY[1.0, 2.0, 3.0], ARRAY[4.0, 5.0, 6.0]);
-- Returns: 27.0
-- Using double arrays
SELECT l2_squared(ARRAY[1.0E0, 2.0E0, 3.0E0], ARRAY[4.0E0, 5.0E0, 6.0E0]);
-- Returns: 27.0

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: zhichenxu-meta (cae80e0)

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D77157252

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D77157252

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D77157252

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D77157252

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D77157252

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D77157252

[steveburnett]

• Please add a release note, even if it's only

== NO RELEASE NOTE ==

• Is documentation needed for this new function? If yes, the release note entry can link to the new documentation: see Formatting in the Release Note Guidelines for examples.

[zhichenxu-meta]

• Please add a release note, even if it's only

== NO RELEASE NOTE ==

• Is documentation needed for this new function? If yes, the release note entry can link to the new documentation: see Formatting in the Release Note Guidelines for examples.
  Thank you and put == NO RELEASE NOTE == for now.

[feilong-liu]

Overall looks good. Leave some comment
And there is a test failure from [Meta Internal-Only Changes Check](https://github.com/prestodb/presto/pull/25409/checks?check_run_id=44781422636)

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D77157252

[zhichenxu-meta]

Overall looks good. Leave some comment And there is a test failure from [Meta Internal-Only Changes Check](https://github.com/prestodb/presto/pull/25409/checks?check_run_id=44781422636)

Feilong, thank you for the review and I have added support for array(double) and follow convention used for other math functions. I will look into the internal checks.

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D77157252

[feilong-liu]

By the way, can you also add the cpp version too to maintain availability of prestissimo engine

[zhichenxu-meta]

By the way, can you also add the cpp version too to maintain availability of prestissimo engine

@feilong-liu I have a separate PR for cpp support, and thank you!

[steveburnett]

• Please add a release note, even if it's only

== NO RELEASE NOTE ==

• Is documentation needed for this new function? If yes, the release note entry can link to the new documentation: see Formatting in the Release Note Guidelines for examples.
  Thank you and put == NO RELEASE NOTE == for now.

Does documentation exist for this function?

[kaikalur]

This function is not handling null values?

[zhichenxu-meta]

This function is not handling null values?

Hi @kaikalur
Dense vectors represented as array(real) or array(double) usually do not contain null elements. I will create a task to add some checks. Thank you!

[kaikalur]

This function is not handling null values?

Hi @kaikalur Dense vectors represented as array(real) or array(double) usually do not contain null elements. I will create a task to add some checks. Thank you!

Presto UDF are general purpose so someone could call it on things with nulls. So we should always keep it general. Also the check is simple in java. Look at the mayHaveNulls() method in Block - error out if that's true for either of the arrays

[zhichenxu-meta]

This function is not handling null values?

Hi @kaikalur Dense vectors represented as array(real) or array(double) usually do not contain null elements. I will create a task to add some checks. Thank you!

Presto UDF are general purpose so someone could call it on things with nulls. So we should always keep it general. Also the check is simple in java. Look at the mayHaveNulls() method in Block - error out if that's true for either of the arrays

Sounds good and thanks for the code pointer.

[kaikalur]

Also if we expect a lot of 0's (I guess for dense vectors that won't happen?) maybe good to short circuit on either element being 0.

[zhichenxu-meta]

Also if we expect a lot of 0's (I guess for dense vectors that won't happen?) maybe good to short circuit on either element being 0.

Good suggestions, and thank you!

Note that we also have c++ implementations of these functions, and the cpp version relies on the FAISS library.