Microsoft has launched ML.NET 2.0, a brand new model of its open supply, cross-platform machine studying framework for .NET. The improve options capabilities for textual content classification and automatic machine studying.
Unveiled November 10, ML.NET 2.0 arrived in tandem with a brand new model of the ML.NET Model Builder, a visible developer device for constructing machine studying fashions for .NET functions. The Mannequin Builder introduces a textual content classification situation that’s powered by the ML.NET Textual content Classification API.
Previewed in June, the Textual content Classification API allows builders to coach customized fashions to categorise uncooked textual content information. The Textual content Classification API makes use of a pre-trained TorchSharp NAS-BERT model from Microsoft Analysis and the developer’s personal information to fine-tune the mannequin. The Mannequin Builder situation helps native coaching on both CPUs or CUDA-compatible GPUs.
Additionally in ML.NET 2.0:
- Binary classification, multiclass classification, and regression fashions utilizing preconfigured automated machine studying pipelines make it simpler to start utilizing machine studying.
- Information preprocessing will be automated utilizing the AutoML Featurizer.
- Builders can select which trainers are used as a part of a coaching course of. Additionally they can select tuning algorithms used to seek out optimum hyperparameters.
- Superior AutoML coaching choices are launched to decide on trainers and select an analysis metric to optimize.
- A sentence similarity API, utilizing the identical underlying TorchSharp NAS-BERT mannequin, calculates a numerical worth representing the similarity of two phrases.
Future plans for ML.NET embrace growth of deep studying protection and emphasizing use of the LightBGM framework for classical machine studying duties similar to regression and classification. The builders behind ML.NET additionally intend to enhance the AutoML API to allow new eventualities and customizations and simplify machine studying workflows.
Copyright © 2022 IDG Communications, Inc.
