AWS Sagemaker 2018- Fully Managed Machine Learning Service

Learn to train a model with Amazon SageMaker using a training job. Specify S3 data, output locations, and compute resources, and explore built-in algorithms, Spark integration, and custom options.

Explain deploying a model on Amazon SageMaker hosting services, including creating a model with S3 artifacts and configuring endpoints with production variants and scalable instances.

Learn to use an Amazon SageMaker notebook instance to explore data, preprocess and transform the training set, and upload it to S3 for training.

Validate machine learning models by evaluating performance and accuracy against business goals using offline and live data. Use production variants and hold-out testing in SageMaker with notebooks and Python SDK.

Explore and preprocess data with an Amazon SageMaker notebook instance and a Jupiter notebook. Train, deploy, and validate models within the fully managed machine learning service.

Explore training a model with AWS SageMaker's built-in algorithm using the NIST handwritten digit dataset, preprocessing data, converting to record IO format, and validating via notebook-based training jobs.

Train and deploy a model using sage maker: create a training job with an s3 data location and training image, then deploy an endpoint and validate with inferences.

Learn how AWS SageMaker's built-in algorithms create models from example data to answer business questions, with discrete classification and quantitative problems, training data formats, and deployment considerations.

Learn how Amazon SageMaker handles data formats for training and inference, including content types, payload structures, CSV and protocol buffers, and transforming training data.

Explore linear learner for supervised classification and regression, training on feature matrices with labeled data, tuning loss and regularization, and evaluating via validation and test data.

Explore factorization machines for high-dimensional sparse data, modeling pairwise feature interactions in classification and regression, and compare with XGBoost's gradient boosting for robust tabular predictions.

Explore image classification with supervised learning using a neural network, trained from scratch or via transfer learning, and learn SageMaker input formats, recordio, image content types, and class probability outputs.

Explore sequence-to-sequence learning in AWS SageMaker, using encoder-decoder neural networks with attention for tasks like machine translation, summarization, and speech-to-text. Understand data formats, training hyperparameters, and inference on SageMaker.

Principal component analysis, an unsupervised method, reduces dimensionality by forming uncorrelated components. In SageMaker, PCA runs in regular or randomized mode for large datasets and outputs the top singular vectors.

Amazon SageMaker's neural topic model is an unsupervised method that discovers topics from document collections using top words, with train, validation, test, and early stopping.

Blazing text, a sage maker algorithm, learns word embeddings from large document collections using skip-gram or cbow, producing word vectors and similarity evaluations.

Learn how to run your own training algorithms and inference code in SageMaker by packaging them in Docker images, configuring training jobs, and supporting file or pipe input modes.

Learn how SageMaker runs your own inference code in a Docker container, using an entry point, exec form, and S3-stored model artifacts to serve inferences and health checks.

Configure automatic scaling for SageMaker production variants with a target-tracking scaling policy driven by CloudWatch metrics. Define minimum and maximum capacity and a cooldown period to control scaling actions.

Configure automatic scaling for a SageMaker endpoint variant using the console or application auto scaling API, setting min and max instances and a target invocations per minute to auto scale.

Define a target tracking scaling policy using predefined or custom metrics, configure it in a json block, and manage cooldowns and disabled activity to control SageMaker variants.

Apply a target tracking scaling policy to a SageMaker production variant via the AWS CLI or Application Auto Scaling API, configuring policy name, resource ID, and invocations per instance.

Edit and delete scaling policies in AWS SageMaker using the management console, CLI, or application auto scaling API, then perform load testing to optimize automatic scaling for endpoints and variants.

Learn how to configure and test AWS SageMaker automatic scaling, manage endpoint updates and scalable targets, and choose between target tracking or step scaling policies for capacity control.

Train and deploy TensorFlow models on Amazon SageMaker using custom TensorFlow code for training and inference, with hands-on examples and tutorials spanning iris data, CNNs, and distributed training.

Launch and manage an MXNet-based training workflow on Amazon SageMaker by creating a notebook, initializing variables, training with fit, and deploying an endpoint for inferences.

Deploy a MXNet model on AWS SageMaker, invoke endpoints to get inferences with the predict method, and train and host models using the SageMaker Spark library and Apache Spark.

Train and host a model with AWS SageMaker and Apache Spark using the Kamin algorithm on a Spark DataFrame of 784-feature 28x28 images from NIST, then infer via endpoint transform.

Learn how SageMaker's Spark pipeline estimators train models using projected features, configure training data with feature and label columns, and manage access with IAM roles and policies.

Understand how AWS SageMaker resources are owned by the AWS account and control access with identity-based policies, cross-account roles, and basic policy elements.

Learn to apply IAM policies to Amazon SageMaker identities, attach permissions to users, groups, or roles, and use AWS managed or custom policies to control SageMaker actions.

Learn how AWS SageMaker uses IAM execution roles with trust policies to grant permissions for notebook instance, training job, and model creation, scoped to S3, DynamoDB, and RDS.

Learn to monitor and diagnose SageMaker deployments with CloudWatch, CloudWatch Logs, and CloudTrail, collecting one-minute metrics, logs, and API call events to ensure reliability and performance.

Understand how CloudTrail log files delivered to an S3 bucket record single requests and actions. Explore create endpoint actions to deploy a train model and create model hosting containers.