ML.NET 1.1 has been released . ML.NET machine learning framework is a cross-platform, designed to allow .NET developers to quickly learn on the phone's browser that allows .NET developers to develop their own models and custom ML injected into the application.
1.1 update highlights include an update ML.NET, and for Visual Studio Model Builder update.
An update of ML.NET
Added support for memory (In-Memory) "image type" in the IDataview
In older versions of ML.NET, each time using the image in the model (for example, using the image of TensorFlow or ONNX scoring model), we need to load images from the drive of the specified file path folders. However ML.NET 1.1 can now be loaded in the image memory and processing them directly.
The new anomaly detection algorithm (preview phase)
In time series NuGet package ( Time Series NuGet Package ) added named SrCnnAnomalyDetection new anomaly detection algorithm ( the Anomaly Detection algorithm ). This algorithm is based on the super-resolution convolution depth network ( Super-Resolution Deep Convolutional Network . ) , One of the advantages is that it does not require any prior training.
To learn more, see this carried out the sample code anomaly detection .
New time series prediction components (preview phase)
This new feature added to the time series NuGet package allows us to achieve based on Singular Spectrum Analysis(SSA) time series forecasting models.
It was named in ML.NET in AdaptiveSingularSpectrumSequenceModeler. When the data has some periodic components of this type are useful for time series forecasting, where causal event, and that they occur at some point in time (or non-occurrence). For example, the sale of different sea (holiday season, sales and weekend schedule, etc.) or to predict the impact of important time component of any other type of data.
Click here to view the relevant sample code .
An update of the Model Builder
This version of the model builder adds support for new programs, and solve many of the problems reported by users.
The introduction of new classification template Issue
This program enables users to add data to the table to classify many types of support. The template using multi-class (multi-class) in classification, can be used to classify data to three or more categories. For example, you can use this template prediction GitHub issue, work orders submitted by the user, as well as e-mail classified into different categories and more scenes.
Improved evaluation step
Evaluation steps (Evaluate step) will now properly display more information about the top model of exploration. This is bad request Most users report repair.
Improved code generation step
To improve the ease of use of the generated code instructions by reference to the project name.