ML.NET 0.6 Released

Microsoft has released the version 0.6 of its ML.NET Machine Learning Framework for .NET developers.

Microsoft has released the version 0.6 of its ML.NET - the Machine Learning framework for .NET developers. This version adds a new and more useful model-building API set, the ability to use pre-trained ONNX Models to provide predictions, and better performance overall.
 
The existing ML.Net API limited the kinds of pipelines you could build and had some restrictions on labeling and scoring the data. The new API supports a wider set of scenarios and closely follows ML principles and naming from other popular ML related frameworks like Apache Spark and Scikit-Learn. This version expands the number of data pipelines that can be used to build machine learning models.
 
According to Microsoft, the new ML.Net API “share a given transform’s execution and transformed data with multiple learners and trainers, or decompose pipelines and add multiple learners.”
 
The new version extends support for predictions derived from the Open Neural Network Exchange format, which is billed as an open platform for interchangeable AI models. ONNX is backed by Microsoft, Amazon Web Services and Facebook.
 
ML.Net 0.6 can use ONNX models trained in multiple frameworks ranging from ML.NET to TensorFlow that can be exported to ONNX. The resulting models can also be used for machine learning applications such as emotion and object recognition.
 
Source: Microsoft 
 
The old ML.Net API will be moved into a legacy namespace so that existing software can continue to use it for the time being.
 
Performance improvements include moving from a legacy “LearningPipeline” API to a new “Estimators” API that serves as an object that learns from data; and optimizing the performance of the PredictionFunction in the new API. Benchmark results revealed a 6,541x speedup performance improvements for use cases like accelerating the parsing of breast cancer data.
 
Use of TensorFlow is made easier for developers to use the open source machine learning library. A new API is added to ML.Net to discover the nodes in the TensorFlow model to help identify the input and output of a TensorFlow model. In ML.NET 0.5 only ‘frozen’ TensorFlow models could be used. Now in ML.NET 0.6, TensorFlow models in the "saved" model format can also be used.
 
The new ML.Net API also uses strong C# types, so that any errors made while designing a pipeline show up earlier in the process and can be linted out.
 
You can read more details and full announcement here.