The Objective

Imagine we aim to create a Streamlit application that generates various random values, such as random integers, upon request. A prototype of such an application could resemble the following:

This Streamlit application is notably straightforward and can be implemented with fewer than 30 lines of code:

# main_page.py

import streamlit as st

from streamlit_random_generator.functions import random_int

from extra_streamlit_tools.utils import init_session_keys

init_session_keys({"randint_upper_limit": 10})

st.title("Generate a random number")

st.number_input(

"Max of random number",

on_change=random_int,

args=(st.session_state["randint_upper_limit"],),

key="randint_upper_limit",

min_value=0,

)

st.button(

"Generate",

on_click=random_int,

args=(st.session_state["randint_upper_limit"],),

)

if "generated_number" in st.session_state:

st.write(st.session_state["generated_number"])

# functions.py

import streamlit as st

import numpy as np

def random_int(x: int):

st.session_state["generated_number"] = np.random.randint(0, x + 1)

To ensure that this application is accessible to everyone, we want to host it online. However, sharing the source code may not always be desirable. To maintain confidentiality, we can store the code in a private GitHub repository.

The Challenge

Streamlit Cloud permits the installation of dependencies solely through a requirements.txt file. An example of a typical requirements.txt could include:

streamlit==1.19.0

extra-streamlit-tools==0.1.0

pandas==1.5.3

But what happens if you've developed your own Python package using a setup.py or pyproject.toml? If this package is published on a library host like PyPI, you can easily include it in your requirements file. However, many users keep their packages on GitHub, where installation is nearly as simple as from PyPI.

To install a package hosted in a GitHub repository, you can prefix the repository link with git+ in your requirements.txt:

streamlit==1.19.0

extra-streamlit-tools==0.1.0

pandas==1.5.3

This approach works seamlessly for public repositories. However, if your repository is private, like our random number generator application, Streamlit will produce an Error 128 during installation due to authentication issues.

The Solution

github_token = ""

Then, incorporate the following code snippet at the beginning of your main_page.py:

import subprocess

import sys

import time

import streamlit as st

try:

import streamlit_random_generator

except ModuleNotFoundError as e:

sleep_time = 30

dependency_warning = st.warning(

f"Installing dependencies, this takes {sleep_time} seconds."

)

subprocess.Popen(

[

],

shell=True,

)

time.sleep(sleep_time)

dependency_warning.empty()

In this setup, during the initial loading of the page, the attempt to import streamlit_random_generator will fail, triggering a ModuleNotFoundError. Consequently, the user will see a message indicating that dependencies are being installed, which will take 30 seconds. This duration can be adjusted based on the number of dependencies.

While the installation occurs, the subprocess module will install your package and its dependencies from the private GitHub repository, using the GitHub token stored in Streamlit secrets. Upon the next reload of main_page.py, the import will succeed, avoiding further installations.

For access to the GitHub repository containing the source code, click [here](#).

Resources

More content available at PlainEnglish.io. Sign up for our free weekly newsletter. Follow us on Twitter, LinkedIn, YouTube, and Discord. If you're interested in scaling your software startup, explore Circuit.