Scientific Plotting

Learning how to plot for publication from Day 1 of research

the-big-book
Author

Raunak Farhaz

Published

October 29, 2025

Chapter 1 - The Hard Way

In research there are 3 sequential processes that go through most of the time. They are enumerated as follows:

  1. Conceptualization
  2. Experimentation
  3. Data Curation
  4. Analysis

The writing of the manuscript is the advertisement and archiving part of the research for the successors to follow and extend it. I am not putting it in the list. I am going to talk about the Analysis part, where a major part goes to visualizing the data extracted from the experiment. Now this is not a tutorial that will answer your questions like - what is the best possible plot to make for a given set of data to have a meaningful illustration of the work? I would suggest to go through any statistics 101 course for that. Here I will talk about the technicalities of drawing a plot with the help computers. I learned it myself the hard way and I am archiving the story here so that you do not need to reinvent your wheel of plotting.

There are many tools as a software to plot beautiful scientific plots but each one of them comes with its pros and cons. I have tried all of them and found out the matplotlib a python package to be the most quickest and simplest to use. Although I will go through briefly why I am not going to discuss about the other tools elaborately.

graph LR
    A[TOOL] --> B(Matplotlib from python)
    A --> C(Plots from julia)
    A --> D(Wolfram Mathematica)
    A --> E(Gnuplot)
    A --> F(Libreoffice)

Julia

Here is the thing Julia is one of my most favourite programming languages. I admire its functionality especially in scientific programming. But it comes with a limitation (although it has been dealt with pretty good in the modern julia versions). The time to the first plot (ttfp) is large. In plain english, it means that the first plot that you generate will be produced slow compared to the second and third plot. The reason here is julia being a just-in-time (JIT) compile language. I will give you a brief example how it looks in terms of syntax and also how the default output looks. For the improvement of ttfp in julia Plots library please have a look into the blogpost by the julia developer himself see https://viralinstruction.com/posts/latency/

using Plots
x = collect(-1:0.1:1)
y = sin.(x)
plot(x,y,xlab="x",ylab="sin(x)")

Pros

  • It is really intuitive and simple.
  • It has a good documentation for most kind of plots see the link https://docs.juliaplots.org/latest/tutorial/
  • It has a huge amount of out of the box styles to be used for your particular purposes

Cons

  • Although the issue of ttfp has been addressed well in v1.9 and even more in v1.10 but it is in news that v1.11 is doing again worse in terms of ttfp compared to the older versions. So you can see it is a bit of a reliability issue.

  • The package for loading dataframes or in plain english simply data from let’s say excel or csv files is called DataFrames. As we speak, I feel the package is not mature enough compared to the equivalent ones in python.

Concluding remarks

If you are using julia for a simulation or an algorithm and you also want to plot some outcomes while running the simulation. Then I would say go for it. Because it would anyways take time so why not just use a functionality that is already given in the programming environment. But if you want it as a standalone data analysis tool then you might feel things are a bit slower.

Wolfram Mathematica

This is meant for a visual and computeresque mathematical tool made by a brilliant physicist Stephen Wolfram in 90s for symbolic computation (which basically means to have a translation of mathematics done in pen and paper to the computer screen). It became a bit of popular tool. But the worst part for the students is it is licensed and not open-source. Recent news although says how wolfram is giving free licenses for academic usage but only the headless environment. So basically the strength of mathematica lies in the notebook environment according to me. But if one does not get to use the notebook then it is a bit of a pity. So I would rather skip this section of syntax completely because the price might be a bit of a hurdle for the students.

Concluding remarks

I also plotted initially many of my research paper plots in mathematica. But sooner I realised scripting with wolfram is a bit of a hurdle especially when you need to plot many of them.

Gnuplot

Gnuplot has nothing to do with GNU (those who are familiar with linux or unix). It is a complete terminal based plotter. It is quite powerful. But the learning curve can be a bit high. I tried it in my masters and then also tried to use it in my PhD. Still it bothers me and always makes me google things which is a bit of a pain. I mean its okay to google things, but not okay to look into google to make even smaller stuffs. But it is definitely powerful but I feel it is outdated. I will skip this too.

Libreoffice

It is useful to make quick data charts to see some instant trends in your data. But for the publication it is less useful and the customization options are limited.

The Winner : Matplotlib

After testing all of these from above I found out that matplotlib is pretty intuitive and easily customizable. And also the packages that you need for data analysis is very rich in python due to its high applicability in the field. The syntax is pretty comfortable and easy to remember. At least you do not need to google all the time for a basic plot. It looks a bit like the following with the default output.

import numpy as np
import matplotlib.pyplot as plt
x=np.arange(-1,1,0.1)
y=np.sin(x)
plt.plot(x,y)
plt.xlabel("x")
plt.ylabel("sin(x)")
plt.show()

So I hope you got a bit of a flavour of how things are. I assume that you might be thinking that the output looks pretty basic. In the next blog post, I will go into the details of matplotlib of how it can be powerful and useful for academic publishing as well as day-to-day data analysis. Then I hope you will also be able to make very amazing and beautiful looking plots.

Chapter 2 - Strong Start

There are innumerous ways to waste time during your first year of Ph.D.

  • How would I produce very beautiful looking plots?
  • What tools are the best?
  • Do you have any script that you use to generate plots?

These questions drove me crazy in my first year and I wasted so much time as there were overwhelming amount of information on these things in internet. But I would have been extremely happy if there was one single workflow that generates

  1. a high resolution image, preferably in PDF
  2. font family that blends well with New Computer Modern font (default font in \(\LaTeX\))
  3. it would be even great that there are already presets for different standard journals
  4. renders mathematical formulas well

Scienceplots will save your day

There is a package in both conda-forge as well as pip repositories called scienceplots. And it solved all of my desires and it will also solve yours too. It basically contains global style settings for matplotlib library so one kind of presets for many standard journals. It also has color-blind friendly palettes. All the functions of matplotlib can be used as it is. So just install and use it. I will go through it one by one through examples. But first let’s install it. There are 3 ways of doing it, do whatever suits you.

  • conda
$ conda install conda-forge::scienceplots
  • pip
$ pip install scienceplots
  • uv
$ uv add scienceplots
Note

uv is a modern python package manager that is written using rust and the installations are blazingly fast compared to conda and pip.

The last python example using matplotlib looked very unprofessional and without anything. Now we will take the same plot and use the scienceplots style backend and generate the plot and we will see how it makes pretty plots effortlessly.

import scienceplots
import numpy as np
import matplotlib.pyplot as plt
plt.style.use(["science"])
x=np.arange(-1,1,0.1)
y=np.sin(x)
plt.plot(x,y)
plt.xlabel("x")
plt.ylabel("sin(x)")
plt.show()

Hence one can see that it has tex fonts rendered within the plot by default “science” style.

Warning

One should be aware that the \(\LaTeX\) should be installed in the machine so that it can use the tex fonts. Otherwise it would throw error.

The amazing part about this package is that one does not need to think about fontfamily, fontsize, color-palletes etc. It kills a lot of time. And at the same time it is customizable and one can change components what they do not like. The next chapter will discuss about the detail of the scienceplots package and other styles. Once you are happy with the plot, export it into a pdf file by the following syntax.

plt.savefig("filename.pdf", bbox_inches='tight')

The bbox_inches='tight' keeps the whitespaces to be minimum. Enjoy your first plot already ready to put in your presentation, papers, notes, and wherever you want.