AWS Sagemaker

https://www.youtube.com/playlist?list=PLhr1KZpdzukcOr_6j_zmSrvYnLUtgqsZz

Amazon SageMaker is a fully managed service that provides every developer and data scientist with the ability to build, train, and deploy machine learning (ML) models quickly. SageMaker removes the heavy lifting from each step of the machine learning process to make it easier to develop high quality models.

Traditional ML development is a complex, expensive, iterative process made even harder because there are no integrated tools for the entire machine learning workflow. You need to stitch together tools and workflows, which is time-consuming and error-prone. SageMaker solves this challenge by providing all of the components used for machine learning in a single toolset so models get to production faster with much less effort and at lower cost.

Many ways to train models on Sagemaker

Built-in Algorithms
Script Mode
Docker container
AWS ML Marketplace
Notebook instance

Amazon SageMaker includes the following features

SageMaker Studio

An integrated machine learning environment where you can build, train, deploy, and analyze your models all in the same application.

SageMaker Model Registry

Versioning, artifact and lineage tracking, approval workflow, and cross account support for deployment of your machine learning models.

SageMaker Projects

Create end-to-end ML solutions with CI/CD by using SageMaker projects.

SageMaker Model Building Pipelines

Create and manage machine learning pipelines integrated directly with SageMaker jobs.

SageMaker ML Lineage Tracking

Track the lineage of machine learning workflows.

SageMaker Data Wrangler

Import, analyze, prepare, and featurize data in SageMaker Studio. You can integrate Data Wrangler into your machine learning workflows to simplify and streamline data pre-processing and feature engineering using little to no coding. You can also add your own Python scripts and transformations to customize your data prep workflow.

SageMaker Feature Store

A centralized store for features and associated metadata so features can be easily discovered and reused. You can create two types of stores, an Online or Offline store. The Online Store can be used for low latency, real-time inference use cases and the Offline Store can be used for training and batch inference.

SageMaker JumpStart

Learn about SageMaker features and capabilities through curated 1-click solutions, example notebooks, and pretrained models that you can deploy. You can also fine-tune the models and deploy them.

SageMaker Clarify

Improve your machine learning models by detecting potential bias and help explain the predictions that models make.

SageMaker Edge Manager

Optimize custom models for edge devices, create and manage fleets and run models with an efficient runtime.

SageMaker Ground Truth

High-quality training datasets by using workers along with machine learning to create labeled datasets. Can reduce labeling cost by 70%

Workers

Mechanical turk workers
Private labeling workforce
Third-party vendors

Label Consolidation

Majority Voting
Probabilities

https://www.labellerr.com/

Amazon Augmented AI

Build the workflows required for human review of ML predictions. Amazon A2I brings human review to all developers, removing the undifferentiated heavy lifting associated with building human review systems or managing large numbers of human reviewers.

SageMaker Studio Notebooks

The next generation of SageMaker notebooks that include AWS Single Sign-On (AWS SSO) integration, fast start-up times, and single-click sharing.

SageMaker Experiments

Experiment management and tracking. You can use the tracked data to reconstruct an experiment, incrementally build on experiments conducted by peers, and trace model lineage for compliance and audit verifications.

SageMaker Debugger

Inspect training parameters and data throughout the training process. Automatically detect and alert users to commonly occurring errors such as parameter values getting too large or small.

SageMaker Autopilot

Users without machine learning knowledge can quickly build classification and regression models.

Good for - classification, regression, some missing values, PCA is ok
Not good for - vision, text, sequence-based, data mostly missing
- You need feature interpretation upfront
- You need pretrained models
Remember
- Try to include as much domain knowledge as you can in the features
- You might have to wait upfront, but it's saving you time in the end
- You will get all the code generated for you

SageMaker Model Monitor

Monitor and analyze models in production (endpoints) to detect data drift and deviations in model quality.

SageMaker Neo

Train machine learning models once, then run anywhere in the cloud and at the edge.

SageMaker Elastic Inference

Speed up the throughput and decrease the latency of getting real-time inferences.

Reinforcement Learning

Maximize the long-term reward that an agent receives as a result of its actions.

Preprocessing

Analyze and preprocess data, tackle feature engineering, and evaluate models.

Batch Transform

Preprocess datasets, run inference when you don't need a persistent endpoint, and associate input records with inferences to assist the interpretation of results.

Built-in Algorithms

Instances

t - tiny

m - memory optimized

c - compute optimized

p - gpu

Best practices

Pick the right size - 5GB default
Think: store on EBS /home/ec2-user/SageMaker
Add or create git repository
Configure security settings
- Encryption
- Root volume access
- Internet access
- VPC Connection
Use a lifecycle config
- Create, start notebook
- Install packages, copy data
- Run in background with '&'
Attach a portion of a GPU for local inference
- Size, version, bandwidth