Microsoft Azure Machine Learning Workbench
I’m a newbie in machine learning and I thought I would try out Azure Machine Learning Workbench.
Machine Learning Workbench is part of a set of tools that Microsoft just released. The tools consist of Azure Machine Learning Experimentation service, the Azure Machine Learning Workbench and the Azure Machine Learning Model Management service. Other tools include Visual Studio Code IDE for building models in CNTK, TensorFlow, Theano, Keras and Caffe2.
The Machine Learning Workbench is a desktop client for Windows and Mac (surprise!). According to Microsoft, it is a “control panel for your development lifecycle and a great way to get started using machine learning.”
Workbench makes it pretty easy to get started learning machine learning. It integrates Jupyter Notebook and and IDEs like Visual Studio Code and PyCharm and allows developers to build models in Python, PySpark and Scala.
You first have to get a Microsoft Azure account. You can get a free account with $200 credit for use with any Azure product for 30 days. After that, you can still get free access to the most popular products for another year and also get 25 always free products. You won’t be charged until you upgrade.
The first thing you do in Workbench is to create a new project. It asks you for a project name, project directory and project description. It also asks for an optional Visualstudio.com GIT Repository URL. You have to fill in the field in the format http://<vsts_account_name>.visualstudio.com/_git/<project_name>.
The full steps are described here https://docs.microsoft.com/en-gb/azure/machine-learning/preview/using-git-ml-project
The Workbench helps you with machine learning by simplifying 3 main steps.
- Simplifying data (see above)
It helps in loading the datasource, cleaning the data by removing unwanted columns, showing you statistics (min, max, median, count etc) about the data by columns, cleaning the data by using functions such as “replace value”, “trim strings” etc. Applying these steps is a breeze. No coding is needed. Once you have completed cleaning, you can export all the steps to python. Much, much easier than having to program all the steps in Python and Jupyter Notebooks. - You can now build a model and in this respects, it is not very dissimilar to building code using Jupyter Notebook. However, once you start to run, you can enter in parameters and record output with the history of the job. This allows you to run many jobs with many different parameters and fine tuning the output of each run by changing the parameters.
- Once you have a trained model. You can now deploy the model as a webservice.
Ok. That’s all the time I have for now. Keep reading and I’ll keep on posting stuff on machine learning.