Name: h2o3-sagemaker
Owner: H2O.ai
Description: Integrating H2O-3 AutoML with Amazon Sagemaker
Created: 2018-03-13 01:56:59.0
Updated: 2018-05-18 18:13:06.0
Pushed: 2018-05-18 18:13:05.0
Homepage:
null
Size: 56
Language: Jupyter Notebook
GitHub Committers
User | Most Recent Commit | # Commits |
Other Committers
User | Email | Most Recent Commit | # Commits |
README
H2O 3 and AWS SageMaker Integration
Proof-of-Concept for integrating H2O-3 AutoML with Amazon SageMaker
Content of Repo, See below for explanation of files:
- README.md
- Dockerfile
- hyperparameters.json
- sample_sagemaker_notebook.ipynb
- automl_scripts
- train
- predictor.py
- nginx.conf
- serve
- wsgi.py
Dockerfile
Used to build the docker image that AWS SageMaker will use for model training purposes
hyperparameters.json
Editable. Contains three nested dictionaries which will be ingested and used during training
- “training” –> will be used to pass along settings such as whether or not to train as a classification problem
- “h2o” –> dictionary of all keyword arguments for h2o.init()
- “aml” –> dictionary of all keyword arguments for H2OAutoML()
sample_sagemaker_notebook
Example of what a jupyter notebook might look like within the AWS SageMaker notebook instance
automl_scripts
The backend code that tells AWS SageMaker what it is expected to do.
files:
- nginx.conf, serve, and wsgi.py do not need to be changed. They are backend code required by AWS SageMaker to build properly
- train and predictor.py likely do not need to be edited. Any arguments you would want to pass are available in hyperparameters.json
To Deploy:
- Create an S3 Bucket with “sagemaker” somewhere in the name. SageMaker will be able to access it in order to pull train/test data
- Create an image repository with AWS ECS
- Build a docker image using
`docker build -t <image name here> -f Dockerfile .
`
- Tag the image appropriately for your AWS ECS repo
`docker tag <image> <new image tag>
`
- Push the image to your AWS ECS repo
`docker push <image>
`
- Follow steps in sample_sagemaker_notebook.ipynb