Oryx 2.2.1

Mostly a bug fix and optimization release. Changes from 2.2.0:

• Much faster ALS startup via faster solver computation, including BLAS acceleration and parallelized computation
• Fixed ALS user/item ID confusion problem
• Removed possibility of ALS user/item ID collision
• Add "logStrength" alternative formulation of ALS model
• Added max-age-model-hours parameter to auto-delete models, not just data
• Don't fail if MODEL-REF refers to deleted model
• Handle case that Kafka topic offsets are outside available range
• Add key-alias property to specify SSL key alias in serving layer
• Various small fixes and optimizations

See some closed issues at https://github.com/OryxProject/oryx/issues?q=milestone%3A2.2.1

Oryx 2.1.3

Mostly a bug fix and optimization release. Changes from 2.1.2:

• Faster ALS startup via faster solver computation
• Fixed ALS user/item ID confusion problem
• Don't fail if MODEL-REF refers to deleted model
• Handle case that Kafka topic offsets are outside available range
• Handle deletes correctly in ALS speed layer
• Various small fixes and optimizations

See some closed issues at https://github.com/OryxProject/oryx/issues?q=milestone%3A2.1.3

Oryx 2.2.0

This is a fairly significant minor release, which has updated minimum requirements:

• Java 8
• Apache Kafka 0.9
• Apache Spark 1.6

Related: 2.2 is required for CDH 5.7.

The move to Java 8 has removed about 1,500 lines of code from the project. Beyond what has been released in the 2.1.x line, this includes minor updates to dependencies as well.

See all closed issues at: https://github.com/OryxProject/oryx/issues?q=milestone%3A2.2.0

Oryx 2.1.2

This is a bug fix release:

• Fix multipart form support
• Fix StackOverflow for ALS with many iterations
• Fix spark-submit quoted arg handling
• Fix some missing dependencies for Serving Layer
• Small optimization to top-N recommendations

See all closed issues at: https://github.com/OryxProject/oryx/issues?q=milestone%3A2.1.2

Oryx 2.0.2

This is a minor bug fix release addressing:

• Fix multipart form support
• Fix JPMML version clash with Spark
• Small optimization to top-N recommendations
• Can now automatically delete old historical data

Oryx 2.1.1

This is a minor bug fix release addressing:

• Version clash in JPMML with Spark 1.5
• Allow conf file to be non-local as intended

All changes: https://github.com/OryxProject/oryx/issues?q=milestone%3A2.1.1

Oryx 2.1.0

This first 2.1.x release requires Spark 1.5, and is compatible with CDH 5.5. It includes some other small enhancements:

• Uses spark-submit for launching the batch and speed layer
• Batch layer's input topic is now optional

This release includes bug fixes from release 2.0.1 as well: https://github.com/OryxProject/oryx/releases/tag/oryx-2.0.1

Oryx 2.0.1

Bug fix release:

• Fixed potential deadlock in batch, speed shutdown
• Fixed class probability output of MLlib-based RDF model
• Restored compatibility with Java serializer

https://github.com/OryxProject/oryx/issues?q=milestone%3A2.0.1

Oryx 2.0.0

• ALS speed layer computes updates in parallel
• Better app example

Closed issues: https://github.com/OryxProject/oryx/issues?q=milestone%3A2.0.0

Oryx 2.0.0 beta 3

Second release candidate for 2.0.0

• k-means endpoints now return distances as Euclidean distance, not squared
• New /recommendWithContext ALS endpoint
• Slightly changed /estimateForAnonymous behavior for consistency with /recommend and /recommendWithContext
• Enabled read-only serving endpoint mode

Closed issues: https://github.com/OryxProject/oryx/issues?q=is%3Aissue+milestone%3A%222.0.0+beta+3%22+is%3Aclosed