ML.NET 0.6 – How is it better than the previous versions?

.Net Engineering Team has officially announced release of ML.NET 0.6. the new framework has been released as the routine monthly release of their cross platform open source framework. There are various exciting features released by the Team in the new framework. The new features promise on improving the experience by proper usage of the machine learning models, performance improvement and more.

Improvement on all Fronts

The team suggests that the brand-new API is better suited for new tasks and code workflow. This, they claim was not possible with the previous LearningPipeline API. Phasing out the current LearningPipeline API is also a part of the plan as the team suggests. The new API according to the developers would expand the list of scenarios its supports. Further, it is line with the ML principles and naming from various other well-known ML related frameworks such as Apache Spark and Scikit – Learn. In the previous update, ML.NET v0.3, the capacity to exporting the ML.NET models to the ONNX -ML format was added. It was done to ensure that the additional execution environments could run swiftly through the model. In the latest version, ML.NET can also utilize the ONNX models to score/predict trained ONNX models running on ONNX standard v1.2. This has been made possible by using a new transformer and runtime for scoring the ONNX models.

Update is simpler to use

The new update has also made it simpler to use TensorFlow models in ML.NET. The team has added an API to identify the nodes in the TensorFlow model to identify the input and output of a TensorFlow model. In ML.NET 0.6, TensorFlow Models which are in the saved formats can also be used. On performance front also, the latest release promises good such as improvement in making single predictions from a trained model. The Legacy LearningPipeline API has been moved to the new Estimator API. Again, optimization has been done in the performance of PredictionFunciton in the new API.

For the latest version, the team also wanted to ensure that it is easier to use ML.NET easier. To achieve this, the Dv type system has been replaced with .NET’s standard type system.


About The Author

This site uses Akismet to reduce spam. Learn how your comment data is processed.