Technology Fridays: Azure ML Workbench Wants to be the Entry Point for your Machine Learning Apps
Welcome to Technology Fridays! Today I would like to continue exploring some of the recent additions to the Azure Machine Learning(ML) stack. specifically, I would like to cover the Azure ML Workbench toolset.
Azure ML Workbench was created to streamline the implementation of machine learning solutions across different frameworks and tools. Sounds trivial? Don’t rush to that conclusion. It turns out that the explosion of development tools and frameworks in the machine learning space has exponentially increased the complexity of the effort required to implement end-to-end solutions. In real machine learning implementations, coordinating a data routines with a Jupyter notebook interactive model with the testing and lifecycle management of your code in a Git repository is fairly complex. Now multiply that complexity by the number of machine learning tools and frameworks in the market and you get a picture of the challenge that Azure ML workbench is trying to solve.
At its core, Azure ML Workbench is a desktop application and a series of a command line tools that attempt to simplify the development lifecycle of machine learning applications. To some extent, Azure ML Workbench can be seen as the entry point and command-control center of your machine learning solution regardless of the underlying technology stack.
Functionally, Azure ML workbench provides a series of capabilities that simplify different areas of the implementation of machine learning programs. For instance, the platform streamlines data preparation tasks by providing a tool that itself leverages machine learning to infer data transformation routines. Similarly, ML Workbench Python SDK enables the execution of those data transformation packages using a very simple programming model.
The integration with other tools and frameworks is the hallmark if Azure ML Workbench. A prime example of this capability is the platform’s support for Jupyter notebooks. Developers using ML Workbench can launch a Jupyter server directly from the application and start creating interactive notebooks. Additionally, ML workbench offers the ability of selecting Jupyter Kernels from pre-configured runtimes that include integration with technologies such as Azure HDInsight. In addition to Jupyter, ML Workbench provides integration with several deep learning frameworks such as TensorFlow, Microsoft Cognitive Toolkit, Chainer and many others.
The interoperability with Git is another key area of innovation of Azure ML Workbench. Any Azure ML Workbench project can be linked to a Git repo and enabling capabilities such as versioning or source control without abandoning the application environment. Not surprisingly, ML Workbench provides a first class integration with VSTS . ML Workbench also streamlines the implementation of machine learning models by suing the Team Data Science Process Templates as well as the integration with different IDEs such as Visual Studio Code or PyCharm which are among’s developers favorites when comes to building machine learning programs.
Azure ML workbench is a very unique solution in the machine learning space. The development toolset of platforms such as H2O.ai or cloud runtimes like DataBricks bears some similarities with ML Workbench but, for the most part, the offering represents a strong differentiator when comes to evaluating the Azure ML stack.