Frameless is a Scala library for working with Spark using more expressive types. It consists of the following modules:

• frameless-dataset for a more strongly typed Dataset/DataFrame API
• frameless-ml for a more strongly typed Spark ML API based on frameless-dataset
• frameless-cats for using Spark's RDD API with cats

Note that while Frameless is still getting off the ground, it is very possible that breaking changes will be made for at least the next few versions.

The Frameless project and contributors support the Typelevel Code of Conduct and want all its associated channels (e.g. GitHub, Discord) to be a safe and friendly environment for contributing and learning.

The compatible versions of Spark and cats are as follows:

* 0.11.0 has broken Spark 3.1.2 and 3.0.1 artifacts published.

† The Spark 4.0.x artifacts (-spark40) are published for Scala 2.13 only and require JDK 17+, since Spark 4 dropped Scala 2.12 and JDK 8/11. The default (unsuffixed) artifacts still target Spark 3.5. † The Spark 4.1.x artifacts (-spark41) are published for Scala 2.13 only and require JDK 17+, since Spark 4 dropped Scala 2.12 and JDK 8/11. The default (unsuffixed) artifacts still target Spark 3.5. † The Spark 4.2.x artifacts (-spark42) are published for Scala 2.13 only and require JDK 17+, since Spark 4 dropped Scala 2.12 and JDK 8/11. The default (unsuffixed) artifacts still target Spark 3.5.

Starting 0.11 we introduced Spark cross published artifacts:

• By default, frameless artifacts depend on the most recent Spark version
• Suffix -spark{major}{minor} is added to artifacts that are released for the previous Spark version(s)

Artifact names examples:

• frameless-dataset (the default Spark 3.5.x dependency)
• frameless-dataset-spark40 (Spark 4.0.x dependency; Scala 2.13 + JDK 17 only)
• frameless-dataset-spark34 (Spark 3.4.x dependency)

Versions 0.5.x and 0.6.x have identical features. The first is compatible with Spark 2.2.1 and the second with 2.3.0.

The only dependency of the frameless-dataset module is on shapeless 2.3.2. Therefore, depending on frameless-dataset, has a minimal overhead on your Spark's application jar. Only the frameless-cats module depends on cats and cats-effect, so if you prefer to work just with Datasets and not with RDDs, you may choose not to depend on frameless-cats.

Frameless intentionally does not have a compile dependency on Spark. This essentially allows you to use any version of Frameless with any version of Spark. The aforementioned table simply provides the versions of Spark we officially compile and test Frameless with, but other versions may probably work as well.

• Spark 3 introduces a new ExpressionEncoder approach, the schema for single value DataFrame's is now "value" not "_1".

Frameless introduces a new Spark API, called TypedDataset. The benefits of using TypedDataset compared to the standard Spark Dataset API are as follows:

• Typesafe columns referencing (e.g., no more runtime errors when accessing non-existing columns)
• Customizable, typesafe encoders (e.g., if a type does not have an encoder, it should not compile)
• Enhanced type signature for built-in functions (e.g., if you apply an arithmetic operation on a non-numeric column, you get a compilation error)
• Typesafe casting and projections

Click here for a detailed comparison of TypedDataset with Spark's Dataset API.

• TypedDataset: Feature Overview
• Typed Spark ML
• Comparing TypedDatasets with Spark's Datasets
• Typed Encoders in Frameless
• Injection: Creating Custom Encoders
• Job[A]
• Using Cats with RDDs
• Proof of Concept: TypedDataFrame

Since the 0.9.x release, Frameless is compiled only against Scala 2.12.x.

To use Frameless in your project add the following in your build.sbt file as needed:

val framelessVersion = "<latest version>"

resolvers ++= Seq(
  // for snapshot artifacts only
  "s01-oss-sonatype" at "https://s01.oss.sonatype.org/content/repositories/snapshots"
)

libraryDependencies ++= List(
  "org.typelevel" %% "frameless-dataset" % framelessVersion,
  "org.typelevel" %% "frameless-ml"      % framelessVersion,
  "org.typelevel" %% "frameless-cats"    % framelessVersion
)

An easy way to bootstrap a Frameless sbt project:

• if you have Giter8 installed then simply:

g8 imarios/frameless.g8

• with sbt >= 0.13.13:

sbt new imarios/frameless.g8

Typing sbt console inside your project will bring up a shell with Frameless and all its dependencies loaded (including Spark).

Feel free to messages us on our discord channel for any issues/questions.

We require at least one sign-off (thumbs-up, +1, or similar) to merge pull requests. The current maintainers (people who can merge pull requests) are:

• adelbertc
• imarios
• kanterov
• non
• OlivierBlanvillain

Frameless contains several property tests. To avoid OutOfMemoryErrors, we tune the default generator sizes. The following environment variables may be set to adjust the size of generated collections in the TypedDataSet suite:

Code is provided under the Apache 2.0 license available at http://opensource.org/licenses/Apache-2.0, as well as in the LICENSE file. This is the same license used as Spark.