sagemaker

The idea is to rebuild AWS Sagemaker Python SDK using R6 classes and paws behind the scenes.

Installation

You can install the development version of sagemaker from GitHub with:

# install.packages("remotes)
remotes::install_github("DyfanJones/sagemaker-r-sdk")

Warning!!!

This repo is in constantly under development and is not currently stable. sagemaker is currently aligning it’s api with sagemaker v2, apologises for any code breaking this causes.

API overview:

This package aims to mimic python’s AWS Sagemaker SDK api, but using R6 and paws

Architecture Design:

sagemaker is a metadata package that contains all methods to interact with Amazon Sagemaker.

• sagemaker.core: Containse core components of sdk for example Session R6 class
• sagemaker.common: Contains common components used throughout sagemaker sdk
• sagemaker.mlcore: Contains core components for machine learning (ML) and amazon developed ML.
• sagemaker.mlframework: Contains ML frameworks developed for Amazon Sagemaker i.e. SKLearn
• sagemaker.workflow: Contains sagemaker pipeline and workflows
• sagemaker.debugger: Contains debugging methods (https://github.com/awslabs/sagemaker-debugger-rulesconfig)

Learn from examples:

Amazon Algorithms:

sagemaker is designed to minic python’s sagemaker sdk. Therefore all examples for python’s sagemaker should be able to accessible.

Examples:

• Targeted Direct Marketing predicts potential customers that are most likely to convert based on customer and aggregate level metrics, using Amazon SageMaker’s implementation of XGBoost.
• XGBoost Tuning shows how to use SageMaker hyperparameter tuning to improve your model fits for the Targeted Direct Marketing task.
• BlazingText Word2Vec generates Word2Vec embeddings from a cleaned text dump of Wikipedia articles using SageMaker’s fast and scalable BlazingText implementation.

R Model Examples:

• R Multivariate Adaptive Regression Splines example over iris data.frame

Note: If a feature hasn’t yet been implemented please feel free to raise a pull request or a ticket

For developers

To keep the package within the CRAN size limit of 5MB. sagemaker is currently using a separate repository (sagemaker-r-test-data) to store R variants of test data stored in sagemaker-python-sdk. sagemaker-r-test-data will only consist of data that can’t be read into R natively i.e. python pickle files. For other test data sagemaker will read it directly from sagemaker-python-sdk.