Skip to content

Latest commit

 

History

History
143 lines (122 loc) · 6.92 KB

release-0.7.md

File metadata and controls

143 lines (122 loc) · 6.92 KB
Raw

ML.NET 0.7 Release Notes

Today we are excited to release ML.NET 0.7, which our algorithms strongly recommend you to try out! This release enables making recommendations with matrix factorization, identifying unusual events with anomaly detection, adding custom transformations to your ML pipeline, and more! We also have a small surprise for those who work in teams that use both .NET and Python. Finally, we wanted to thank the many new contributors to the project since the last release!

Installation

ML.NET supports Windows, MacOS, and Linux. See supported OS versions of .NET Core 2.0 for more details.

You can install ML.NET NuGet from the CLI using:

dotnet add package Microsoft.ML

From package manager:

Install-Package Microsoft.ML

Release Notes

Below are some of the highlights from this release.

  • Added Matrix factorization for recommendation problems (#1263)

    • Matrix factorization (MF) is a common approach to recommendations when you have data on how users rated items in your catalog. For example, you might know how users rated some movies and want to recommend which other movies they are likely to watch next.
    • ML.NET's MF uses LIBMF.
    • Example usage of MF can be found here. The example is general but you can imagine that the matrix rows correspond to users, matrix columns correspond to movies, and matrix values correspond to ratings. This matrix would be quite sparse as users have only rated a small subset of the catalog.
    • Note: ML.NET 0.3 included Field-Aware Factorization Machines (FFM) as a learner for binary classification. FFM is a generalization of MF, but there are a few differences:
      • FFM enables taking advantage of other information beyond the rating a user assigns to an item (e.g. movie genre, movie release date, user profile).
      • FFM is currently limited to binary classification (the ratings needs to be converted to 0 or 1), whereas MF solves a regression problem (the ratings can be continuous numbers).
      • If the only information available is the user-item ratings, MF is likely to be significantly faster than FFM.
      • A more in-depth discussion can be found here.

  • Enabled anomaly detection scenarios (#1254)

    • Anomaly detection enables identifying unusual values or events. It is used in scenarios such as fraud detection (identifying suspicious credit card transactions) and server monitoring (identifying unusual activity).
    • This release includes the following anomaly detection techniques: SSAChangePointDetector, SSASpikeDetector, IidChangePointDetector, and IidSpikeDetector.
    • Example usage can be found here.

  • Enabled using ML.NET in Windows x86 apps (#1008)

    • ML.NET can now be used in x86 apps.
    • Some components that are based on external dependencies (e.g. TensorFlow) will not be available in x86. Please open an issue on GitHub for discussion if this blocks you.

  • Added the CustomMappingEstimator for custom data transformations #1406

    • ML.NET has a wide variety of data transformations for pre-processing and featurizing data (e.g. processing text, images, categorical features, etc.).
    • However, there might be application-specific transformations that would be useful to do within an ML.NET pipeline (as opposed to as a pre-processing step). For example, calculating cosine similarity between two text columns (after featurization) or something as simple as creating a new column that adds the values in two other columns.
    • An example of the CustomMappingEstimator can be found here.

  • Consolidated several API concepts in MLContext #1252

    • MLContext replaces LocalEnvironment and ConsoleEnvironment but also includes properties for ML tasks like BinaryClassification/Regression, various transforms/trainers, and evaluation. More information can be found in #1098.
    • Example usage can be found here.

  • Open sourced NimbusML: experimental Python bindings for ML.NET.

    • NimbusML makes it easy for data scientists to train models in Python and hand them off to .NET developers to include in their apps and services using ML.NET.
    • NimbusML components easily integrate into scikit-learn pipelines.
    • Note that NimbusML is an experimental project without the same support level as ML.NET.

Acknowledgements

Shoutout to dzban2137, beneyal, pkulikov, amiteshenoy, DAXaholic, Racing5372, ThePiranha, helloguo, elbruno, harshsaver, f1x3d, rauhs, nihitb06, nandaleite, timitoc, feiyun0112, Pielgrin, malik97160, Niladri24dutta, suhailsinghbains, terop, Matei13, JorgeAndd, and the ML.NET team for their contributions as part of this release!