Data Science with Python on Azure AI

Machine Learning Introduction

Before ML can be applied the key concepts of machine learning need to be discussed.

• Why using machine learning
• Supervised versus unsupervised learning
• Machine learning methodology CRISP-DM
• Data preparation
• Train, validation and test sets
• Classification, regression and clustering

AI Services

Pretrained models are an easy way to get started building intelligent applications. These models are pre-trained for generic tasks such as image or voice recognition, text analysis and much more. Some models can be further fine-tuned to meet specific needs (transfer learning) This module presents the different AI services (formerly called Azure Cognitive Services) in Azure, split up in domains such as vision, speech, decision, etc.

• AI Services
• Azure OpenAI
• Speech service
• Language service
• Computer vision
• Customizing AI services
• LAB: Setup AI services and train a custom vision model

Azure OpenAI

Large Language models (LLMs) such as Chat-GPT have proven to be useful in a wide variety of scenarios. By providing predefined prompts, you can configure these models to help you implement chat bots, answer questions for staff and customers, summarize documents etc. Microsoft offers this service as Azure OpenAI, integrating it with the Azure platform.

• Introducing OpenAI
• Azure OpenAI Studio
• Tokenizer
• Choosing the right Deployment
• Completion versus Chat
• Deploying models as a web app
• Creating,storing and using Embeddings

Using Azure AI Services in Python

This module shows how the Azure AI services can be used from within Python code, or using a command line interface (CLI).

• Azure AI services API
• Using Azure AI Services CLI
• Calling Azure AI Services from Python
• LAB: Programming on top of Azure AI Services

Machine Learning with scikit-learn

Many business problems can be tackled by basic machine learning techniques. In this module the most common machine learning approaches such as linear regression and random forests are used with Python. We also pay attention to model inspection.

• Machine learning specific data preprocessing: normalization, standardization, one-hot encoding
• Classification using decision trees, logistic regression and support vector machines
• Model tuning: working with hyper-parameters
• Building regression models with linear regression, SVM and Neural networks
• Unsupervised learning: Clustering
• LAB: Classification and Regression with scikit-learn

Azure Machine Learning Services

Machine learning on a local machine and a small dataset is one thing, running this on larger datasets or more CPU-hungry techniques can become a challenge. Another problem is deploying your model: How can we easily call the resulting model from within other applications? Azure Machine Learning Services helps answering these questions.

• Azure ML service overview
• Create a ML service workspace
• Setting up computes and datastores
• Creating and querying experiments
• Deploying and using models
• Creating and registering images
• Deploy images as web services
• LAB: Building ML models in Azure Machine Learning

Automated Machine Learning

Part of the machine learning process can be automated. Azure Automated machine learning provides a web portal and a Python API to automate data preprocessing and model training.

• Building and deploying Automated ML models via the portal
• Calling Automated ML from Python
• LAB: Working with Automated ML

Azure Machine Learning Pipelines

Machine learning pipelines allow you to build parameterized, reusable pipelines with data cleansing, model training and evaluation steps. These can be made from Python code or with a graphical designer in the Azure ML portal.

• Azure ML pipelines
• Using the Azure ML Designer
• Creating pipelines in Python
• Executing Azure ML pipelines
• LAB: Building and executing Azure ML pipelines

Getting started with Deep Learning

From all the machine learning techniques there is one that gets popular for more challenging problems: Multiple layers of neural networks, better known as deep learning. For problems such as image recognition, speech understanding etc. this is currently the way to go. But it's from a mathematical point of view a very challenging technique. In this module the basics of deep learning are introduced.

• From Neural networks to Deep learning
• Overview of deep learning frameworks
• Getting started with the Keras framework