Image Classification (Preprocess, Train & Deploy) in AWS Sagemaker

A simple end to end Image Classification (preprocessing, training, deployment) using AWS Sagemaker.

NOTE : Using pipeline is in different repo.

Steps to Preprocess, Train & Deploy

Upload data file in .zip (data source https://www.kaggle.com/datasets/puneet6060/intel-image-classification) to S3. s3://mlops-tutorials/sagemaker-mlops1/intel/
Create a repo in AWS CodeCommit for DVC
Run setup-git-dvc notebook
Run data-reprocessing notebook
Run train notebook
Run deployAndPredict notebook
Save JSON file in SM_MODEL_DIR directory

# save eval metrics in json file
    trainer.test(model, datamodule)
    eval_metrics = { k : v.tolist() for k, v in trainer.callback_metrics.items() }
    print(eval_metrics)
    with open((sm_model_dir/"eval_metrics.json"), "w") as jfile:
        json.dump(eval_metrics, jfile)

Check Cloud Watch logs

Example of Inference

Top 5 predictions for each image.

Create Custom Docker image from AWS DLC

Create Custom Docker Image for Sagemaker Jobs using AWS Deep Learning Containers as base.

Why should I extend a SageMaker container?

Install additional dependencies. (E.g. I want to install a specific Python library, that the current SageMaker containers don't install.)
Configure your environment. (E.g. I want to add an environment variable to my container.)

Permissions

SageMakerFullAccess and AmazonEC2ContainerRegistryFullAccess to create the custom docker

how to package a PyTorch container, extending the SageMaker PyTorch container by extending the SageMaker PyTorch container we can utilize the existing training and hosting solution made to work on SageMaker.
AWS DLC (Deep Learning Containers), choose one of these as base https://github.com/aws/deep-learning-containers/blob/master/available_images.md

e.g. 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:1.12.1-cpu-py38-ubuntu20.04-sagemaker

Spin up a EC2 instance
Create a requirements.txt with required packages and versions
Create a dockerfile file with below contents

#Take base AWS DLC container
FROM 763104351884.dkr.ecr.ap-southeast-2.amazonaws.com/pytorch-training:1.12.1-cpu-py38-ubuntu20.04-sagemaker

COPY requirements.txt requirements.txt

RUN pip3 install -r requirements.txt \
    && rm -rf /root/.cache/pip

Create a repo in ECR, copy build commands
From EC2 terminal LOGIN : () aws ecr get-login-password --region ap-southeast-2 | docker login --username AWS --password-stdin 763104351884.dkr.ecr.ap-southeast-2.amazonaws.com
Then Build DOCKER docker build -t custom-pytorch-cpu-sagemaker .
After the build is completed, tag your image so you can push the image to this repository: docker tag custom-pytorch-cpu-sagemaker:latest 536176424191.dkr.ecr.ap-southeast-2.amazonaws.com/custom-pytorch-cpu-sagemaker:latest
Then again login to aws ecr get-login-password --region ap-southeast-2 | docker login --username AWS --password-stdin 536176424191.dkr.ecr.ap-southeast-2.amazonaws.com
Run the following command to push this image to your newly created AWS repository: docker push 536176424191.dkr.ecr.ap-southeast-2.amazonaws.com/custom-pytorch-cpu-sagemaker:latest

Refer Building AWS Deep Learning Containers Custom Images

Tensorboard Logs

Go to terminal and check for tensorboard, if not present then install pip install tensorboard
Run tensorboard --logdir <s3 URI of logs> e.g.

tensorboard --logdir s3://sagemaker-ap-southeast-2-536176424191/sagemaker-intel-classification-logs/training-intel-dataset-2022-12-15-15-44-51-564/tensorboard-output/

Upload to TensorBoard.dev

Uploading the TensorBoard logs will give you a URL that can be shared with anyone.
Uploaded TensorBoards are public, so do not upload sensitive data.
The uploader will exit when the entire logdir has uploaded. (This is what the --one_shot flag specifies.)

tensorboard dev upload --logdir s3://sagemaker-ap-southeast-2-536176424191/sagemaker-intel-classification-logs/training-intel-dataset-2022-12-15-15-44-51-564/tensorboard-output/ \
--name "Intel image Classification" \
--description "git link here" \
--one_shot

Link to Tensorflow Dev :

https://tensorboard.dev/experiment/Ry8FY2MvTr2CaM15k2ikwA/#scalars

References :

https://docs.aws.amazon.com/sagemaker/latest/dg/build-your-own-processing-container.html
https://github.com/aws/amazon-sagemaker-examples/blob/main/advanced_functionality/pytorch_extending_our_containers/pytorch_extending_our_containers.ipynb
AWS DLC (Deep Learning Containers), choose one of these as base https://github.com/aws/deep-learning-containers/blob/master/available_images.md
Custome Docker Image https://docs.aws.amazon.com/deep-learning-containers/latest/devguide/deep-learning-containers-custom-images.html
Setup for HTTPS users using Git credentials https://docs.aws.amazon.com/codecommit/latest/userguide/setting-up-gc.html?icmpid=docs_acc_console_connect_np

aiplaybookin/aws-sagemaker-intel-classification