Index

Domain 1: Data Preparation for Machine Learning

Let's get started with Domain 1, which covers data preparation for machine learning. For this domain, we'll continue to talk about machine learning and the first two steps of the machine learning lifecycle. Remember, the first step of the machine learning lifecycle is collecting a lot of quality data to input into your algorithm. The second step is to process the data to ensure the quality of the data is high, ensure there is no missing data, ensure there is no erroneous data, and to manipulate the data to be as efficient as possible. Domain 1 is broken into three task statements that we will discuss over the next few lessons. Task statement 1.1, ingest and store data. Task statement 1.2, transform data and perform feature engineering. Task statement 1.3, ensure data integrity and prepare data for modeling. For the first task statement to ingest and store data, you'll need to understand how to ingest data and extract data from the different AWS storage services based on the storage access patterns, data formats and structures, cost, and performance. This task statement also covers troubleshooting and debugging data ingestion and storage issues involving scalability and capacity. For the second task statement to transform data and perform feature engineering, ensure you understand AWS services and tools and the techniques to clean and transform your data. You should also understand the AWS tools to explore, visualize, and transform the features of your data. This includes feature engineering and encoding techniques in AWS services to validate and label your data. For the third task statement, to ensure data integrity and prepare data for modeling, ensure you understand techniques such as classification, anonymization, and masking your data. You should also understand the AWS services that can help you to encrypt your data. This task statement will focus on data integrity and validating data quality, identifying and mitigating bias, and reducing the bias prediction. Over the next few videos, I will address each task statement individually, breaking down the knowledge and skills expected of you to be successful. Let's get started evaluating your readiness for the exam, in the next lesson where we will cover the first task statement from Domain 1.

Domain 2: Model Training and Evaluation

Let's get started with Domain 2, which covers machine learning model development. In Domain 1, we talked about the first two steps of the machine learning lifecycle to collect data and to process data. This domain covers the third step, which is to use the selected algorithm to train the model, and to generate a model you can use to derive value, and the fourth step, which is to train the model. And for the well-architected machine learning lifecycle, this domain covers step four, which is model development. Domain 2 is broken into three task statements that we'll cover over the next few lessons. Task statement 2.1: Choose a modeling approach, task statement 2.2: train and refine models, task statement 2.3: analyze model performance. For the first task statement to choose a modeling approach, you will need to understand the capabilities and appropriate use cases of machine learning algorithms to solve problems, including SageMaker's built-in algorithms, how to use the AWS artificial intelligence services to solve specific problems, and how to consider interpretability during the model or algorithm selection. For the second task statement, to train and refine models, ensure you understand the training process, methods to reduce the model training time, factors that influence the model size, benefits of regularization techniques, hyperparameters tuning techniques, how model hyperparameters affect model performance, and how to integrate models built outside of SageMaker. For the third task statement, to analyze model performance, you'll need to know model evaluation metrics and techniques, and methods to create performance baselines, convergent issues, and metrics available in SageMaker Clarify to gain insights into your machine learning training data and models. Over the next few videos, I will address each task statement individually breaking down the knowledge and skills expected of you to be successful. Let's get started evaluating your readiness for the exam, in the next lesson, where we will cover the first task statement from Domain 2.

Domain 3: Model Deployment and Operations

Let's get started with Domain 3, which covers the deployment and orchestration of machine learning workloads. This domain covers the fifth step of the machine learning lifecycle, which is to host the model. For the well-architected machine learning lifecycle, Domain 3 covers step five, which is deployment. Domain 3 is broken into three task statements that we'll discuss over the next few lessons. Task statement 3.1, select deployment infrastructure based on existing architectures and requirements. Task statement 3.2, create and script infrastructure based on existing architectures and requirements. And task statement 3.3, use automated orchestration tools to set up continuous integration and continuous delivery pipelines. For the first task statement, to select deployment infrastructure based on existing architectures and requirements, you will need to understand deployment best practices, including model and endpoint requirements for deployment endpoints, methods to optimize models on edge devices, and methods to serve machine learning models in real time and in batches. This task statement also covers provisioning compute resources, which we began talking about under Domain 2. For the second task statement, to create and script infrastructure based on existing architectures and requirements, we will dive deeper into the provision resources, scaling policies, infrastructure as code, containerization, and how to use SageMaker endpoint auto scaling policies. For the third task statement, to use automated orchestration tools to set up continuous integration and continuous delivery pipelines, ensure you understand CI/CD principles and how they fit into your machine learning workflows. Ensure you understand how to add automation and integration of your data ingestion with orchestration services, how to use version control systems, and how your code repository works when a pipeline is invoked. Under this task statement, you also need to know deployment strategies, which is also under task statement 3.1, but this task statement dives deeper into rollback actions. Over the next few videos, I'll address each task statement individually, breaking down the knowledge and skills expected of you to be successful. Let's get started evaluating your readiness for the exam in the next video, where we will cover the first task statement from Domain 3.

Domain 4: Machine Learning Security

Let's get started with Domain 4, which covers machine learning solution monitoring, maintenance, and security. Throughout this course, we have been talking about the first five steps of the machine learning lifecycle and the well-architected machine learning lifecycle. Domain 4 covers the sixth step of both lifecycles, which is to make inference and model monitoring. Domain 4 is broken into three task statements that we'll discuss over the next few lessons. Task statement 4.1: Monitor model inference. Task statement 4.2: Monitor and optimize infrastructure costs. And task statement 4.3: Secure AWS resources. For the first task statement, to monitor model inference, you will need to understand drift in your machine learning models, techniques to monitor data quality and model performance, and the design principles for machine learning lenses that are relevant to monitoring. The second task statement, to monitor and optimize infrastructure costs, is broken into two parts, optimizing your infrastructure and optimizing your costs. For the first part of this task statement, ensure you know how to identify key performance metrics for your machine learning infrastructure. Also ensure you know the AWS monitoring and observability tools that you can use to troubleshoot latency and performance issues. For the second part of this task statement, which is cost optimization, ensure you know the differences between instance types and how the different instance types affect performance. And ensure you know how to use the AWS cost analysis tools, cost tracking, and allocation techniques. For the third task statement, to secure AWS resources, you will need to know how to use IAM roles, policies, and groups, bucket policies, and Amazon SageMaker Role Manager to control access to your AWS services. This task statement also covers SageMaker security and compliance features, network access controls to your machine learning resources, and security best practices for your CI/CD pipelines. Over the next few videos, I will address each task statement individually, breaking down the knowledge and skills expected of you to be successful. Let's get started evaluating your readiness for the exam in the next video, where we will cover the first task statement from Domain 4.