FREE AI Classes: Machine Learning with Azure

Introduction

Introduction

We are very glad to have you on this call. We welcome everyone from the University very warmly. As you can testify, this is going to be another awesome and great session to learn how to master the use of Azure for machine learning. This will make machine learning building very easy for us.

Since Monday, we have been learning, and you can now do something with what you have learned. Feel free to drop your experience in the chat for others to learn from.

Important Information

There is an event on Nvidia Trustworthy AI that we would like everyone to join. It’s happening today, and my colleague has dropped the link to register. If you are available at 2:30, you can join.

The aim of today's session is not just to build AI solutions but to build responsible and transparent AI solutions. So, without further ado, let's introduce our tutor, Emmanuel.

Session Overview

Tutor: Emmanuel

Day 1 - Azure Subscription and Resources

On Monday, we started with Azure machine learning, covering what it takes to manage a machine learning workload on the Azure platform. We discussed how to create an Azure subscription and a resource group.

Day 2 - Azure Machine Learning Workspace

On Tuesday, we delved into Azure Machine Learning Studio, familiarizing ourselves with the functionalities available. We learned how to upload data and create a machine learning workspace.

Day 3 - Experimentation and Optimization

Today, we focus on:

• Using Automated Machine Learning (AutoML)
• Tracking Machine Learning Models with MLflow
• Optimizing Model Training and Hyperparameter Tuning
• Using ML Pipelines

Using AutoML

AutoML simplifies the machine learning process by automating data cleaning, feature engineering, and model training. This is particularly useful when working with complex data sets. AutoML in Azure supports various machine learning algorithms like logistic regression, decision trees, and XGBoost.

Tracking Models with MLflow

Tracking models involves logging and monitoring model performance, and this is crucial for machine learning operations. MLflow is a tool that helps in tracking and organizing model training runs.

Optimizing Model Training

Optimization can be achieved by refactoring code, setting parameter tuning, and automating job submission.

Using ML Pipelines

ML Pipelines allow for the organization of a sequence of steps involving data processing, model training, and predictions as a single workflow. This helps in automating and scaling machine learning operations.

Practical Implementation

AutoML Implementation

• Data Cleaning: Use AutoML to automatically clean and prepare data.
• Feature Engineering and Model Selection: AutoML selects the best features and models based on the provided metrics.

MLflow Implementation

• Tracking Parameters: Use MLflow to log parameters, metrics, and artifacts for model comparison.
• Using Auto Log and Custom Log: Log essential details automatically or manually.

Hyperparameter Tuning

• Sampling Methods: Implement grid search or random sampling to find the best hyperparameters.
• Early Stopping and Termination: Set policies to stop training rules to save resources.

Creating Pipelines

• Component Creation: Use Jupyter Notebook or VS Code to create components.
• Pipeline Definition: Define pipelines using YAML configuration files.
• Automation: Automate tasks using job schedules for efficient resource usage and management.

Keyword

• Azure
• Machine Learning
• AutoML
• MLflow
• Hyperparameter Tuning
• Pipelines
• Data Cleaning
• Feature Engineering
• Model Tracking
• Job Automation

FAQ

Q: What is AutoML in Azure? A: AutoML automates the processing of machine learning tasks including data cleaning, feature selection, and model training to simplify the machine learning workflow.

Q: How can I track machine learning experiments in Azure? A: You can use MLflow for tracking machine learning models, logging parameters, metrics, and artifacts to monitor and compare model performance.

Q: How do I optimize model training in Azure? A: You can optimize model training by refactoring code, setting parameter tuning, and automating job submissions.

Q: What are the benefits of using ML Pipelines in Azure? A: ML Pipelines help in organizing, automating, and scaling sequences of machine learning tasks, which reduces manual intervention and increases efficiency.

Q: How can I create a component in Azure Machine Learning? A: Create components by defining scripts and YAML files that specify metadata, interfaces, commands, and environments, and register them in the Azure Machine Learning workspace.

Q: What methods can be used for hyperparameter tuning in Azure? A: Grid search and random sampling methods can be used for hyperparameter tuning to find the best configurations for model training.

Conclusion

Today, we have learned about using Azure for machine learning, focusing on automated machine learning, tracking and optimizing models, and building machine learning pipelines. With these skills, you can make your machine learning projects more efficient and scalable.

Do watch out for upcoming events and hackathons. Thank you for joining, and have a great day!