GitHub - alejohz/ABICHALLENGE_ALEJANDRO-HENAO: Repo containing all documentation of the ML deployment test.

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 27 Commits
.github/workflows		.github/workflows
__pycache__		__pycache__
.gitignore		.gitignore
Dockerfile		Dockerfile
README		README
ahenao_web_app.py		ahenao_web_app.py
docker-compose.yml		docker-compose.yml
helloworld.py		helloworld.py
master_script.sh		master_script.sh
read_upload.py		read_upload.py
requirements.txt		requirements.txt
script.py		script.py
sm_model.py		sm_model.py
testing_sm_model.ipynb		testing_sm_model.ipynb

Repository files navigation

Developed by: Alejandro Henao
Date: 01-May-2022

Repo composed of python files for the development of a web app using Streamlit using a pretrained model inside S3 and deploying an
Endpoint with Sagemaker for on demand REQUESTS of the API.


Proposed work if i had more time:
- Develop streamlit app inside EC2
- Use the secrets Github functionality for AWS credentials
- Upload Docker image to Dockerhub for dependencies, this would help to upload to ECR and be able to run app inside EC2.


Conclusions:
Developing the streamlit app inside EC2 would prove to be a better solution due to not running on local, however EC2 instances are very high cost.

AWS Sagemaker is pretty slow and costly for on demand, probably AWS Lambda might be a better solution.

Streamlit is pretty limmitted and has only linear workflow conditions, Flask could prove to be a more robust micro web framework solution.

XGBoost is incredibly good for tabular data for very low training time and hyperparameter tuning.

NOTE: CODE IS NOT FUNCTIONAL, access keys are deactivated.