Train a Deep Learning Model on SageMaker

A repo to train a Custom Deep Learning Model with Amazon SageMaker

1. How to create buckets for input and output

aws s3 mb s3://sagemaker-train-data-hacene
aws s3 mb s3://sagemaker-output-data-hacene

2. How to download and uplaod the dataset to S3

kaggle datasets download -d alxmamaev/flowers-recognition
unzip flowers-recognition.zip
aws s3 sync flowers s3://sagemaker-train-data-hacene/flowers

3. Create an ECR Repository

account=$(aws sts get-caller-identity --query Account --output text --profile=hacene)

region=$(aws configure get region)

aws ecr get-login-password --region $region| docker login --username AWS --password-stdin $account.dkr.ecr.$region.amazonaws.com

aws ecr create-repository --repository-name train_sagemaker

# delete the repository if it exists
aws ecr delete-repository --repository-name train_sagemaker --force --region $region

4. Build and push the Image

./build_and_push.sh

5. Launch the Training Job

 aws sagemaker create-training-job --cli-input-json file://training-config.json