Summer 2024 DoC Prof's Project

About

This project explores some different features of Streamlit & Flask to build a more comprehensive app.

Current Project Components

Currently, there are three major components:

• Streamlit App (in the ./app directory)
• Flask REST api (in the ./api directory)
• MySQL setup files (in the ./database directory)

Getting Started

1. Clone the repo to your computer.
2. Set up the .env file in the api folder based on the .env.template file.
3. Start the docker containers.

Handling User Role Access and Control

In most applications, when a user logs in, they assume a particular role. For instance, when one logs in to a stock price prediction app, they may be a single investor, a portfolio manager, or a corporate executive (of a publicly traded company). Each of those roles will likely present some similar features as well as some different features when compared to the other roles. So, how do you accomplish this in Streamlit? This is sometimes called Role-based Access Control, or RBAC for short.

The code in this project demonstrates how to implement a simple RBAC system in Streamlit but without actually using user authentication (usernames and passwords). The Streamlit pages from the original template repo and split up among 2 roles - Political Strategist and USAID Worker, and a System Administrator role is used for any sort of system tasks such as re-training the ML model, etc. It also demonstrates how to deploy an ML model.

Wrapping your head around this will take a little time and exploration of this code base. Some highlights are below.

Getting Started with the RBAC

1. We need to turn off the standard panel of links on the left side of the Streamlit app. This is done through the .streamlit/config.toml file. So check that out. We are turning it off so we can control directly what links are shown.
2. Then I created a new python module in modules/nav.py. When you look at the file, you will se that there are functions for basically each page of the application. The st.sidebar.page_link(...) adds a single link to the sidebar. We have a separate function for each page so that we can organize the links/pages by role.
3. Next, check out the Home.py file. Notice that there are 3 buttons added to the page and when one is clicked, it redirects via st.switch_page(...) to that Roles Home page in ./pages. But before the redirect, I set a few different variables in the Streamlit session_state object to track role, first name of the user, and that the user is now authenticated.
4. Notice near the top of Home.py and all other pages, there is a call to SideBarLinks(...) from the nav module. This is the function that will use the role set in session_state to determine what links to show the user in the sidebar.
5. I reorganized the pages by Role. Pages that start with a 0 are related to the Political Strategist role. Pages that start with a 1 are related to the USAID worker role. And, pages that start with a 2 are related to The System Administrator role.

Deploying An ML Model

Note: This project only contains the infrastructure for a hypothetical ML model.

1. Convert your Jupyter Notebook code for an ML model to a purely python script. You can include the training and testing functionality as well as the prediction functionality. You don't need to include data cleaning, though.
2. In api/backend, notice the addition of ml_models module. In this folder, I've put a sample (read fake) ML model in model01.py. The predict function will be called by the Flask REST API to perform real-time prediction based upon model parameter values that are stored in the database. Important: you cannot hard code the model parameter weights directly in the prediction function. tl;dr - take some time to look over the code in model01.py.
3. The prediction route for the REST API is in backend/customers/customer_routes.py. Basically, it accepts two URL parameters and passes them to the prediction function in the models module. The prediction function packages up the value(s) it receives from the model's predict function and send its back to Streamlit as JSON.
4. Back in streamlit, check out pages/11_Prediction.py. What I do on this page is create two numeric input fields. When the button is pressed, it makes a request to the REST API /c/prediction/.../... function and passes the values from the two inputs as URL parameters. It gets back the results from the route and displays them. Nothing fancy here.