## Intro to deployment <!--v--> ### Architecturing (web) applications  <!-- .element: height="50%" width="50%" --> <!--v--> ### Communicating between applications [Application Programming Interface](https://glossary.cncf.io/application-programming-interface/) > Une API (application programming interface ou « interface de programmation d’application ») est une interface logicielle qui permet de « connecter » un logiciel ou un service à un autre logiciel ou service afin d’échanger des données et des fonctionnalités. <!--v--> ### API  <!-- .element: height="50%" width="50%" --> <!--v--> ### REST API Representational state transfer (REST)  <!-- .element: height="50%" width="50%" --> <!--v--> ### Microservices vs "monoliths"  <!-- .element: height="50%" width="50%" --> PS: [Microservices are hard](https://dwmkerr.com/the-death-of-microservice-madness-in-2018/) <!--v--> ### Multi applications & docker  <!-- .element: height="50%" width="50%" --> <!--v--> ### How does it relate to me ? <!--v--> ### Hands-On - **Backend** How to expose an ML model to a community of users through a web app - **Frontend** How to build a companion app to interact with your model in an ergonomic fashion - **Deployment** How to deploy both applications on a single instance <!--v--> In the data science workflow,  <!-- .element: height="50%" width="50%" --> <!--v--> I have an **awesome** ML model  <!-- .element: height="50%" width="50%" --> <!--v--> (Just kidding)  <!-- .element: height="50%" width="50%" --> <!--v--> I want to deploy it on the cloud for other to use  <!-- .element: height="50%" width="50%" --> <!--v--> Today we will do it by hand  <!-- .element: height="50%" width="50%" --> <!--v--> Other methods for ML Model packaging behind a web server  - cog : https://github.com/replicate/cog - pesto : https://github.com/AirbusDefenceAndSpace/pesto - litserve : https://github.com/Lightning-AI/LitServe <!--v--> Interaction with user ? We use CURL 👎 ```bash curl -X POST "http://my-instance/predict" \ -H "accept: application/json" \ -H "Content-Type: application/json" \ -d "{\"model\":\"string\",\"image\":\"...\"}" ``` <!--v--> Interaction with user ? We use CURL 👎  <!-- .element: height="30%" width="30%" --> <!--v--> Interaction with users ? 👍  <!-- .element: height="50%" width="50%" --> <!--v--> Webapp builder for data scientists  <!-- .element: height="50%" width="50%" --> [you've seen it before](http://supaerodatascience.github.io/DE/1_4_be.html#6-lets-discover-streamlit) <!--v--> Webapp builder for data scientists - [streamlit](https://streamlit.io/) - [gradio](https://gradio.app/) <!--v--> Let's build it ! - A model behind a Restful API, packaged in a docker - A frontend using streamlit, packaged in a docker - Deploy it on Google Cloud Platform using GCE & docker-compose - Send it to your friends ! <!--v--> In reality, it's much more complex...  <!-- .element: height="50%" width="50%" --> <!--v--> How to scale deployment ? - [CS229 - Class by Chip Huyen](https://docs.google.com/presentation/d/1U_zKs19VLJKnGE02JDRnzxJ8lgeVF22WSZ_GrA646fY/edit#slide=id.p) - [CS229 - Deployment with Ray Serve](https://github.com/anyscale/academy/blob/main/ray-serve/e2e/tutorial.ipynb) - https://docs.ray.io/en/latest/serve/develop-and-deploy.html <!--v--> Some links - https://github.com/EthicalML/awesome-production-machine-learning - Machine Learning System Designs https://stanford-cs329s.github.io/
