News
ML.NET Machine Learning Framework Update Addresses Usability and Stability
After hitting version 1.0 last month, Microsoft's open source, cross-platform machine learning framework ML.NET has received its first update, adding functionality and addressing developer concerns about usability and stability in the Model Builder component.
ML.NET 1.1, consisting of a Visual Studio UI tool called Model Builder and a command-line interface (CLI), works with AutoML to help C# and F# coders create machine learning models.
By adding machine learning functionality to .NET applications, they can use available data to make predictions and address use cases such as:
- Classification/categorization
- Regression/predict continuous values
- Anomaly detection
- Recommendations
- Sentiment analysis
- Object detection
New in ML.NET 1.1, as detailed in a June 11 blog post, are:
- Preview of support for in-memory "image type" in IDataview: Developers can now load in-memory images and process them directly instead of having to specify file paths for images stored on a hard drive.
- Preview of Anomaly Detection algorithm: A new Anomaly Detection algorithm named SrCnnAnomalyDetection has been added to the Time Series NuGet package, without requiring any prior training for use.
- Preview of Time Series Forecasting components: Also added to the Time Series NuGet package, a new component provides series forecasting predictions useful "when your data has some kind of periodic component where events have a causal relationship and they happen (or miss to happen) in some point of time."
- An internal update to TensorFlow, now using version 1.13.1 (formerly 1.12.0)
- Assorted bug fixes
All of the above and more are detailed in the release notes.
The Model Builder components also received several updates, including: a new issue classification template (for classifying tabular data into many classes); improved evaluate and code generation steps; and fixes for customer feedback issues concerning installation errors, usability and stability.
About the Author
David Ramel is an editor and writer at Converge 360.