[SPARK-12626] MLlib 2.0 Roadmap

• Type: Umbrella
• Status: Resolved
• Priority: Blocker
• Resolution: Done
• Fix Version/s: 2.0.0
• Component/s: ML, MLlib
• Labels: roadmap
• Target Version/s: 2.0.0
• Assignee: Xiangrui Meng
• Reporter: Joseph K. Bradley
• Created: 04/Jan/16 20:15
• Updated: 26/May/16 20:44
• Resolved: 29/Apr/16 21:40

Description

This is a master list for MLlib improvements we plan to have in Spark 2.0. Please view this list as a wish list rather than a concrete plan, because we don't have an accurate estimate of available resources. Due to limited review bandwidth, features appearing on this list will get higher priority during code review. But feel free to suggest new items to the list in comments. We are experimenting with this process. Your feedback would be greatly appreciated.

Instructions

For contributors:

• Please read https://cwiki.apache.org/confluence/display/SPARK/Contributing+to+Spark carefully. Code style, documentation, and unit tests are important.
• If you are a first-time Spark contributor, please always start with a starter task rather than a medium/big feature. Based on our experience, mixing the development process with a big feature usually causes long delay in code review.
• Never work silently. Let everyone know on the corresponding JIRA page when you start working on some features. This is to avoid duplicate work. For small features, you don't need to wait to get JIRA assigned.
• For medium/big features or features with dependencies, please get assigned first before coding and keep the ETA updated on the JIRA. If there exist no activity on the JIRA page for a certain amount of time, the JIRA should be released for other contributors.
• Do not claim multiple (>3) JIRAs at the same time. Try to finish them one after another.
• Remember to add the `@Since("2.0.0")` annotation to new public APIs.
• Please review others' PRs (https://spark-prs.appspot.com/#mllib). Code review greatly helps to improve others' code as well as yours.

For committers:

• Try to break down big features into small and specific JIRA tasks and link them properly.
• Add a "starter" label to starter tasks.
• Put a rough estimate for medium/big features and track the progress.
• If you start reviewing a PR, please add yourself to the Shepherd field on JIRA.
• If the code looks good to you, please comment "LGTM". For non-trivial PRs, please ping a maintainer to make a final pass.
• After merging a PR, create and link JIRAs for Python, example code, and documentation if applicable.

Roadmap (WIP)

This is NOT a complete list of MLlib JIRAs for 2.0. We only include umbrella JIRAs and high-level tasks.

Major efforts in this release:

• `spark.ml`: Achieve feature parity for the `spark.ml` API, relative to the `spark.mllib` API. This includes the Python API.
• (SPARK-13944) Linear algebra: Separate out the linear algebra library as a standalone project without a Spark dependency to simplify production deployment.
• Pipelines API: Complete critical improvements to the Pipelines API
• New features: As usual, we expect to expand the feature set of MLlib. However, we will prioritize API parity over new features. New algorithms should be written for `spark.ml`, not `spark.mllib`.

Algorithms and performance

• iteratively re-weighted least squares (IRLS) for GLMs (SPARK-9835)
• estimator interface for GLMs (SPARK-12811)
• extended support for GLM model families and link functions in SparkR (SPARK-12566)
• improved model summaries and stats via IRLS (SPARK-9837)

Additional (maybe lower priority):

• robust linear regression with Huber loss (SPARK-3181)
• vector-free L-BFGS (SPARK-10078)
• tree partition by features (SPARK-3717)
• local linear algebra (SPARK-6442)
• weighted instance support (SPARK-9610)
  • random forest (SPARK-9478)
  • GBT (SPARK-9612)
• locality sensitive hashing (LSH) (SPARK-5992)
• deep learning (SPARK-5575)
  • autoencoder (SPARK-10408)
  • restricted Boltzmann machine (RBM) (SPARK-4251)
  • convolutional neural network (stretch)
• factorization machine (SPARK-7008)
• distributed LU decomposition (SPARK-8514)

Statistics

• bivariate statistics as UDAFs (SPARK-10385)
• R-like statistics for GLMs (SPARK-9835)
• sketch algorithms (cross listed) : approximate quantiles (SPARK-6761), count-min sketch (SPARK-6763), Bloom filter (SPARK-12818)

Pipeline API

• pipeline persistence (SPARK-6725)
  • trees (SPARK-11888)
  • RFormula (SPARK-11891)
  • MLC (SPARK-11871)
  • PySpark (SPARK-11939) --> This is now ready for people to take up subtasks!
• ML attribute API improvements (SPARK-8515)
• predict single instance (SPARK-10413)
• test Kaggle datasets (SPARK-9941)

There may be other design improvement efforts for Pipelines, to be listed here soon. See (SPARK-5874) for a list of possibilities.

Model persistence

• PMML export
  • naive Bayes (SPARK-8546)
  • decision tree (SPARK-8542)
• model save/load
  • FPGrowth (SPARK-6724)
  • PrefixSpan (SPARK-10386)
• code generation
  • decision tree and tree ensembles (SPARK-10387)

Data sources

• public dataset loader (SPARK-10388)

Python API for ML

The main goal of Python API is to have feature parity with Scala/Java API. You can find a complete list here. The tasks fall into two major categories:

• Pipeline persistence in PySpark (SPARK-11939)
• Python API for missing methods (SPARK-11937)
• Python API for new algorithms. Committers should create a JIRA for the Python API after merging a public feature in Scala/Java.

SparkR API for ML

• support more families and link functions in SparkR::glm (SPARK-12566)
• model summary with R-like statistics for GLMs (SPARK-9837)
• support more algorithms (k-means (SPARK-13011), survival analysis (SPARK-13010), etc.)

Documentation

• re-organize user guide (SPARK-8517)
• make example code testable in user guide (SPARK-11337)
• @Since versions in spark.ml, pyspark.mllib, and pyspark.ml (SPARK-7751)
• Fix param format in pydoc (SPARK-11219)

Issue Links

• relates to: SPARK-10324 MLlib 1.6 Roadmap
• relates to: SPARK-15581 MLlib 2.1 Roadmap