Machine Learning at AWS

AWS Service Overview

General Purpose Machine Learning Services

• Amazon Elastic Inference - attach GPU acceleration to Amazon instances: https://aws.amazon.com/machine-learning/elastic-inference/
• Amazon SageMaker - general purpose machine learning: https://aws.amazon.com/sagemaker/
• AWS Deep Learning AMI (DLAMI) - Linux with Deep Learning Hardware (AMI = Amazon Machine Image)
  • https://aws.amazon.com/machine-learning/amis/
  • https://docs.aws.amazon.com/dlami/latest/devguide/what-is-dlami.html

Text and Speech Services

• Amazon Polly - text to speech service (blackbox): https://aws.amazon.com/polly/
• Amazon Comprehend - natural language processing (NLP) service (blackbox): https://aws.amazon.com/comprehend/
• Amazon Lex - Chat like Alexa (speech recognition and natural language understanding): https://aws.amazon.com/lex/
• Amazon Textract - extract text and data from scanned documents (blackbox): https://aws.amazon.com/textract/
• Amazon Transcribe - speech to text (blackbox): https://aws.amazon.com/transcribe/
• Amazon Translate - neural machine translation service (blackbox): https://aws.amazon.com/translate/

Image and Video Services

• Amazon Rekognition - image and video analysis (blackbox): https://aws.amazon.com/rekognition/

Toy Services

• AWS DeepLens - deep learning video camera (toy): https://aws.amazon.com/deeplens/
• AWS DeepRacer - autonomous deep learning 1/18th scale car (toy): https://aws.amazon.com/deepracer/

Other Services

• Amazon Forecast - time-series forecasting service (blackbox): https://aws.amazon.com/forecast/
• Amazon Personalize - personalization and recommendation (blackbox): https://aws.amazon.com/personalize/
• Amazon SageMaker Ground Truth - manual labeling tools: https://aws.amazon.com/sagemaker/groundtruth/
• Amazon SageMaker Neo - compile models for speed and low memory footprint: https://aws.amazon.com/sagemaker/neo/

SageMaker Links

• Amazon SageMaker 10-Minute Tutorial
• Amazon SageMaker Technical deep-dive
• Learn fundamentals and deep dive into Amazon SageMaker algorithmsLearn more
• SageMaker - Developer Guide - Get Started

Python API

• Boto 3 Documentation

Services

• P3 (V100 GPU): https://aws.amazon.com/de/ec2/instance-types/p3/
• P2 (K80 GPU): https://aws.amazon.com/de/ec2/instance-types/p2/
• G3 (M60 GPU): https://aws.amazon.com/de/ec2/instance-types/g3/

Performance Experiment with EC2 and LightGBM

Given:

• LightGBM Classification
• x_train.shape (13788428, 30)
• x_val.shape (2433253, 30)
• y_train.shape (13788428,)
• y_val.shape (2433253,)
• num_classes: 401
• num_boost_round: 4

Results:

• c5.24xlarge, 4,656 $/h, 96 vCPUs, 2nd Gen Intel Xeon Platinum 8275CL, 3.6 GHz, 192 GiB RAM, time: 512 s
• m5.8xlarge, 1,84 $/h, 32 vCPUs, Intel Xeon Platinum 8175, 3.1 GHz, 128 GiB RAM, time: 574 s
• r5.4xlarge, 1,216 $/h, 16 vCPUs, Intel Xeon Platinum 8175, 3.1 GHz, 128 GiB RAM, time: 763 s
• other server, 4 (8) CPUs, Xeon W-2123, 3.6 GHz, 128 GiB RAM, time: 856 s

Connect

• Convert Key: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/putty.html