Deep Berlin AI For Good Hackathon 2020

Train Your Model with AWS

Preparation

Ideally, you set up your AWS account 3-4 days before the hackathon and increase SageMaker Instance Limits as described below. This is specifically important if you want to leverage a larger number of GPU-based cloud instances to run your training jobs.

Step 1: Create and Activate Your AWS Account

https://aws.amazon.com/premiumsupport/knowledge-center/create-and-activate-aws-account/

Step 2: Create an S3 Bucket

How to Create an S3 Bucket

Step 3: Choose a Region

Review the Regions supported by Amazon SageMaker.

Step 4: Manage/Increase SageMaker Instance Limits (see detailed instructions here)

Review the Default Limits for Amazon SageMaker Service Limits
Request a Limit Increase if needed via the AWS Support Center
Review the SageMaker Instance Pricing for an overview of supported instance types

Let's get started!

Step 5: Apply Your AWS Credits (If Applicable)

https://aws.amazon.com/awscredits/

Step 6: Create a SageMaker Notebook Instance

Navigate to Amazon SageMaker (https://console.aws.amazon.com/sagemaker/) (make sure you're in the region of your choice shown in the top right menu of your screen)
Navigate to Noteboook Instances in the left side menu
Click on Create notebook instance
- Choose a name for your notebook instance
- Select an Instance Type
- You might want to increase the volume size of the locally attached disk (i.e. to 250GB or 500GB)

Select Create a New IAM Role
- Select Specific S3 Bucket and type in the name of the S3 bucket you created earlier
- Click Create role

In Git repositories you can choose to clone a public Git repo to this notebook instance
- Just provide the public Git repo URL

You can leave everything else to default, and hit Create notebook instance
Your notebook instance is now being created (this can take 2-3min)

Once the instance shows Status InService you can connect to it via Open Jupyter or Open JupyterLab.
This opens up your notebook environment with the cloned Git repo ready.

Note: You can also connect a Private GitHub or GitLab repo, just follow these instructions.

Step 7: Clone any other GitHub Repo

If you want to clone another Git repo:

Open a New Terminal
Clone the Git repo using the following command (we are using this repo as an example again):

cd ~/SageMaker
git clone https://github.com/data-science-on-aws/ai4good-hackathon.git

Step 8: Train a Model Using Your SageMaker Notebook Instance

Sample TensorFlow Notebook using Distributed TensorFlow and SageMaker.
Sample PyTorch Notebook using Distributed PyTorch and SageMaker.
Sample MXNet Notebook using Distributed MXNet and SageMaker.
To adapt a custom training script to SageMaker, please follow these instructions.

shamshad/ai4good-hackathon