Right now, one of the fascinating subfields of Data Science is Automated Machine Learning (AutoML). It sounds fantastic for those unfamiliar with machine learning, but it concerns present Data Scientists. The media presentation of AutoML suggests that technology has the potential to drastically transform the way we produce models by removing the need for Data Scientists. In principle, utilizing AutoML to automate the process entirely is a brilliant idea, but it introduces several opportunities for bias and misunderstanding in practice.

Machine learning model training can be a time-consuming process. Automated Machine Learning (AutoML) makes identifying the best strategy for your circumstance and dataset easier. ML.NET is an open-source, cross-platform machine learning framework for .NET developers that allows custom machine learning to be integrated into .NET applications. Microsoft changed the AutoML implementation in its Model Builder and ML.NET CLI tools based on Microsoft Research’s Neural Network Intelligence (NNI) and Fast and Lightweight AutoML (FLAML) technology last year. These improvements provided various advantages and improvements over the previous system, including:

NNI / FLAML AutoML implementations have been incorporated into the ML .NET framework, allowing use from a code-first perspective.

To get started with the AutoML API, install the latest pre-release version of Microsoft.ML and Microsoft .ML.Auto NuGet packages using the ML.NET daily feed.

https://pkgs.dev.azure.com/dnceng/public/_packaging/MachineLearning/nuget/v3/index.json

The experiment API includes a set of AutoML defaults, making it easier to add to a training pipeline. The dataPrepPipeline in this code snippet is a sequence of transforms to get the data into the proper format for training. The AutoML components required to train a regression model are added to that pipeline.

The same idea holds for other supported cases, such as categorization. When building an experiment using the training pipeline, one may choose the length of the training, the training and validation sets, and the evaluation measure they are optimizing. After setting up the pipeline and experiment, call the Run function to begin training.

Model Builder and the ML .NET CLI have both been updated. Two of them are as follows:

The trainer for time-series forecasting in ML .NET is straightforward to select because there is only one choice, ForecastBySsa. The hardest part is deciding which elements to analyze. Such as the time frame to investigate and how far into the future to plan. Finding the correct parameters is an iterative process that is great for AutoML.

Clean and representative data improves the performance of the model. As a result, data analysis, cleansing, and preparation for training is a crucial stages in the machine learning workflow. A few years back, we introduced the DataFrame type to.NET as a preview in Microsoft.Data.Analysis NuGet package. The DataFrame is still in testing. Microsoft understands how important it is for developers to have the tools needed to execute data cleaning and processing tasks, so we have started collecting and prioritizing feedback to address existing reliability and developer experience concerns. The contributions are being combined into a GitHub issue.

Machine Learning Operations (MLOps) is a machine learning lifecycle equivalent to DevOps. It comprises features such as model deployment and administration and data tracking, which aids in producing machine learning models. Microsoft continually strives for ways to improve the ML .NET experience. These are some updates brought in by Microsoft ib their ML .NET framework, which will help developers better their workflow.

Source: https://devblogs.microsoft.com/dotnet/whats-new-with-mldotnet-automl/

No thanks, I’m not interested!