Authoring Tip: Efficiently Managing Generated Latex Assets


When writing scientific reports and research I usually include assets such as tables and images that are included in the final report using code. Generated assets are often dependent on numerical results that may change between runs depending on the version of the code used or random noise. The majority of technical authors manually upload their plots to Overleaf and manually enter their tables into their final .tex document. This is a huge time sink and disincentivizes you to continue iterating on your project.

This tutorial shows you how to efficiently manage your generated assets using Overleaf and GitHub.

Project structure

Let's start out by creating a new project. After creating your project, initialize the Overleaf GitHub sync feature. I used to not use this because it is a premium feature. Don't make the same mistake. Value your time and sanity over saving $8 a month.

initialize an Overleaf GitHub sync project

When you initialize GitHub sync, Overleaf will produce a GitHub repo for your project. Idiomatically, this repo should only include your paper code. To work around this, this repo should be treated as a submodule of your primary repo.

Create another GitHub repo. Following this, clone your repo, and add you Overleaf repo as a submodule:

git submodule add '$YOUR_GITHUB_NAME/$OVERLEAF_REPO'

To make managing the submodule easier I update my global git config to pull changes by default:

git config --global submodule.recurse true

For demonstration purposes lets create a directory that will generate some assets:

mkdir -p experiments/artifacts

Cool, your project is setup now. You have a GitHub project that will host experiments that generate latex tables and an Overleaf repo as a submodule.

Generating and syncing assets

For demonstration purposes lets generate two assets:

  • a line plot
  • a Latex table

Line Plot

For demonstration purposes we will produce a simple line plot. Enter the following code in a file called experiments/

# experiments/
import matplotlib.pyplot as plt
import numpy as np

x = np.linspace(-1, 1, 50)
y = x

plt.plot(x, y)
line plot

Latex Table

Let's also produce a simple table using Pandas. Let's add the following code to experiments/

import numpy as np
import pandas as pd
data = {
  'Metric': ['Accuracy', 'Recall'],
  'Value': ['$24.333\pm4.005$', '$94.333\pm3.999$'],
  # including this to demonstrate how float formatting works
  'Pi': [np.pi, np.pi]

df = pd.DataFrame(data)
result = df.to_latex(
    caption='Numerical Results',

with open('artifacts/demo-table.tex', 'w') as f:
demo table

df.to_latex() offers an easy way to produce latex tables.

Add the artifacts to Overleaf

First, run both files:

cd experiments

Next, let's add these artifacts to Overleaf:

cp artifacts/* ../test-paper/artifacts/
cd ../test-paper
git commit -am "Update artifacts"
git push origin master

Finally, pull the changes from GitHub:

Pull changes from GitHub

Including the assets in your Latex document

Let's go through the best way to include these assets in your Latex document.


For figures it's pretty straight forward. You simply create a figure and reference the resulting image:

line plot

Easy as that.


Including tables is also trivial. Recall that we saved our table produced by Pandas to artifacts/demo-table.tex. You'll include the table as follows:

demo table

Input is a low level macro that will basically act as though the text in demo-table.tex was hosted in your Overleaf document.

Updating Artifacts

Whenever you want to update the artifacts, you will just run the following:

cp artifacts/* ../test-paper/artifacts/
cd ../test-paper
git commit -am "Update artifacts"
git push origin master

Then, go to Overleaf and pull the changes from GitHub:

Pull changes from GitHub

Following this your updates will be rendered in the overleaf document. Super quick and easy!

Hope this helps, happy writing!


Other Posts