Custom workflows¶
Note
We’re constantly adding examples to this page, so please check back soon for more. Or, if you have a request or a workflow you’d like to share, please either open an issue or suggest an edit to this page by clicking the GitHub link at the top.
Default figure generation¶
By default, the workflow defined in the Snakefile
looks like this:
# User config
configfile: "showyourwork.yml"
# Import the showyourwork module
module showyourwork:
snakefile:
"showyourwork/workflow/Snakefile"
config:
config
# Use all default rules
use rule * from showyourwork
The default behavior in this workflow is to infer figure dependencies based on
the figure labels in the tex file.
The following block in ms.tex
\begin{figure}
\begin{centering}
\includegraphics{figures/mandelbrot.pdf}
\caption{The Mandelbrot set.}
% This label tells showyourwork that the script `figures/mandelbrot.py'
% generates the PDF file included above
\label{fig:mandelbrot}
\end{centering}
\end{figure}
tells showyourwork
to execute a script called mandelbrot.py
in the src/figures
directory to generated figures/mandelbrot.pdf
.
To change, supplement, or override this behavior, read on!
Multi-panel figures¶
It is possible to include multiple figures within a figure
environment, provided
they are all generated by the same script:
\begin{figure}[ht!]
\begin{centering}
\includegraphics[width=0.4\linewidth]{figures/koch1.pdf}
\includegraphics[width=0.4\linewidth]{figures/koch2.pdf}
\caption{
Two Koch snowflakes.
}
% This label tells showyourwork that the script `figures/koch.py'
% generates the two PDF files included above
\label{fig:koch}
\end{centering}
\end{figure}
If you would like to include figures generated from different scripts in the
same figure
environment, you’ll have to provide a custom rule (see below).
One script, multiple figures¶
Conversely, we can also have different figure environments, all of which include
figure files generated from the same script.
If you follow the usual convention, this would result in duplicated labels, since
these figure environments would share the same label (determined only by the name of the script
that generated them).
To get around this, showyourwork
supports adding tags to the end of figure labels
to make them unique.
\begin{figure}[ht!]
\begin{centering}
\includegraphics[width=0.4\linewidth]{figures/mandelbrot.pdf}
\caption{
This figure was generated by the script \texttt{mandelbrot.py}
and is labeled \texttt{fig:mandelbrot:original}.
}
\label{fig:mandelbrot:original}
\end{centering}
\end{figure}
\begin{figure}[ht!]
\begin{centering}
\includegraphics[width=0.4\linewidth]{figures/mandelbrot_red.pdf}
\caption{
This figure was generated by the script \texttt{mandelbrot.py}
and is labeled \texttt{fig:mandelbrot:red}.
}
\label{fig:mandelbrot:red}
\end{centering}
\end{figure}
In the example above, the script mandelbrot.py
generates two PDFs, which
are displayed in separate figure environments. We label them fig:mandelbrot:original
and fig:mandelbrot:red
to make them unique; showyourwork
ignores everything
after the second colon, and understands that both figure environments correspond to
the same figure script (mandelbrot.py
).
Static figures¶
It is also possible to commit the figure PDF/PNG/SVG/etc directly and tell showyourwork
not to
try to produce it programmatically. Simply place the figure in the src/static
directory:
\begin{figure}[ht!]
\begin{centering}
\includegraphics[width=0.4\linewidth]{static/broccoli.pdf}
\caption{
A photo of some broccoli.
}
% The fact that the figure is in the static directory tells
% showyourwork not to look for a script that generates this figure
\label{fig:broccoli}
\end{centering}
\end{figure}
Script dependencies¶
Sometimes we would like to tell showyourwork
about script dependencies, such as
when our figure script imports something from a locally-hosted script or package.
We can do this by specifying a dependency in the configuration
file showyourwork.yml
:
# Tell showyourwork that `src/figures/my_figure.py`
# depends on `src/figures/utils/helper_script.py`
dependencies:
src/figures/my_figure.py:
- src/figures/utils/helper_script.py
Note that all paths are relative to the root of your repository.
Dataset dependencies¶
If you have a dataset hosted on Zenodo, showyourwork
can automatically
download it for you and link to it in the corresponding figure caption in the article PDF.
All you have to do is specify the name of the dataset and its Zenodo record ID in the config
file showyourwork.yml
:
# Tell showyourwork that `src/figures/fibonacci.py`
# requires the file `src/data/fibonacci.dat` to run
dependencies:
src/figures/fibonacci.py:
- src/data/fibonacci.dat
# Tell showyourwork where to find ``src/data/fibonacci.dat``
zenodo:
- src/data/fibonacci.dat:
id: 5187276
The YAML snippet above tells showyourwork
that the script src/figures/fibonacci.py
requires the dataset src/data/fibonacci.dat
in order to run.
It also tells showyourwork
that this dataset can be downloaded from Zenodo, and that
it has the record ID 5187276
. Specifically, that means the file lives at the URL
https://zenodo.org/record/5187276 and can be downloaded by running
curl https://zenodo.org/record/5187276/files/fibonacci.dat
Note that if this dataset is a tarball, you’ll have to untar it within fibonacci.py
, or
specify a custom rule in the Snakefile
(see below).
Alternatively, you can manually specify how to download a dataset dependency.
This is useful if, e.g., it’s hosted somewhere other than Zenodo, or if you
need to do some post-processing (like unzipping) before running your figure
script. To do that, simply don’t specify the entry in the zenodo
section of your
showyourwork.yml
file:
dependencies:
src/figures/fibonacci.py:
- src/data/fibonacci.dat
and instead create a custom rule in the Snakefile
:
# Custom rule to download a dataset
rule fibonacci:
output:
"src/figures/fibonacci.dat"
shell:
"curl https://zenodo.org/record/5187276/files/fibonacci.dat --output {output[0]}"
Note that this approach will not automatically add a dataset link to your figure caption.
Simulation dependencies¶
Quite often you may have a figure that is very computationally expensive to run. An example is a posterior distribution plot for an MCMC run, or a plot of an expensive fluid dynamical simulation. If the runtime is more than a few tens of minutes (on a single machine), you probably don’t want to run it on GitHub Actions, even if you rely on showyourwork caching. One way around this is to run the simulation, upload the results to Zenodo (via the workflow discussed above), and treat that as a static “dataset” on which your figure depends. The downside, however, is that your workflow is no longer fully reproducible, since it depends on the result of a black-box simulation.
To address this, showyourwork
supports dynamic rules that can alternate between running the simulation
and uploading to Zenodo (when running on a local machine), and downloading the simulation from Zenodo (when
running on GitHub Actions). This can be achieved by specifying additional instructions in the
showyourwork.yml
file:
dependencies:
src/figures/my_figure.py:
- src/data/simulation.dat
zenodo:
- src/data/simulation.dat:
script: src/figures/run_simulation.py
sandbox: false
token_name: ZENODO_TOKEN
title: Simulation results
description: >-
This is the result of a very expensive simulation.
Here is some text describing the simulation in detail,
how it was generated, and how to use the dataset.
creators:
- Luger, Rodrigo
There’s a lot going on in this example, so let’s break it down piece by piece.
First, we’re telling showyourwork
that the figure script src/figures/my_figure.py
requires the result of some expensive simulation, stored in the data file src/data/simulation.dat
.
Then, under zenodo:
, instead of specifying the id:
of the dataset, we instead explicitly tell
showyourwork
how to generate it with a script
key. Specifically, we specify the Python
script
that runs the simulation.
Note
showyourwork
executes the Python script
from within the directory containing it.
In this example, the simulation script is executed from within the src/figures/
directory,
so it must save the simulation file as ../data/simulation.dat
, since that’s where
showyourwork
expects to see it based on the config file.
The next several instructions tell showyourwork
how to upload the results of the simulation
to Zenodo. The title
, description
, and creators
keys should be self-explanatory: they
will show up in the metadata section of the Zenodo deposit.
The sandbox
key is a boolean flag telling showyourwork
whether to use the Zenodo Sandbox
service (the default is False); this is useful for testing and debugging, and should be disabled
once you release your code/paper.
Finally, since showyourwork
will upload the results of the simulation to Zenodo, it needs your
credentials to access the API. So, in order for this all to work, you need to do three things:
If you haven’t done this already, create a Zenodo account and generate a personal access token. Make sure to give it at least
deposit:actions
anddeposit:write
scopes, and store it somewhere safe.To give
showyourwork
access to Zenodo from your local machine, assign your token to an environment variable calledZENODO_TOKEN
. I export mine from within my.zshrc
or.bashrc
config file so that it’s always available in all terminals.To give
showyourwork
access to Zenodo from GitHub Actions, create a repository secret in your GitHub repository calledZENODO_TOKEN
and set its value equal to your Zenodo token.
Warning
Never include your personal access tokens in any files committed to GitHub!
Now you should be all set. Make sure to run your expensive simulation locally before pushing your changes to GitHub – otherwise GitHub Actions won’t find the file on Zenodo, and the build will fail.
If all goes well, you should see an icon pop up next to the corresponding figure caption with a link to the record on Zenodo for your simulation results.
Dependency tarballs¶
showyourwork
also supports the upload/download of Zenodo tarballs. Consider the following showyourwork.yml
file:
dependencies:
src/figures/my_figure.py:
- src/data/results_0.dat
- src/data/results_1.dat
- src/data/results_2.dat
- src/data/results_3.dat
- src/data/results_4.dat
- src/data/results_5.dat
- src/data/results_6.dat
- src/data/results_7.dat
- src/data/results_8.dat
- src/data/results_9.dat
zenodo:
- src/data/results.tar.gz:
script: src/data/run_simulation.py
sandbox: false
token_name: ZENODO_TOKEN
title: Random numbers
description: >-
This is a collection of ten datasets, each containing
ten iid zero-mean, unit-variance random numbers. These
are used in an example of the showyourwork open source
scientific article workflow.
creators:
- Luger, Rodrigo
contents:
- src/data/results_0.dat
- src/data/results_1.dat
- src/data/results_2.dat
- src/data/results_3.dat
- src/data/results_4.dat
- src/data/results_5.dat
- src/data/results_6.dat
- src/data/results_7.dat
- src/data/results_8.dat
- src/data/results_9.dat
This is similar to the previous example, except this time the figure script
depends on a large number of simulation result files. By specifying a contents
key under a zenodo
entry, we can instruct showyourwork
to generate the
tarball results.tar.gz
out of those contents and upload it to Zenodo.
We then list all of those individual files as dependencies of the figure script.
This example works in the same way as above – the simulation is only ever
run locally. So again, make sure to run it before pushing your changes
to GitHub – otherwise GitHub Actions won’t find the tarball on Zenodo,
and the build will fail.
By the way, there’s a handy feature of the YAML syntax that can save us some repetition: anchors and aliases. It’s a handy way of defining and re-using chunks of metadata. You can read more about them here. We can use the anchor/alias syntax to re-write the YAML file above as
dependencies:
src/figures/my_figure.py: &results
- src/data/results_0.dat
- src/data/results_1.dat
- src/data/results_2.dat
- src/data/results_3.dat
- src/data/results_4.dat
- src/data/results_5.dat
- src/data/results_6.dat
- src/data/results_7.dat
- src/data/results_8.dat
- src/data/results_9.dat
zenodo:
- src/data/results.tar.gz:
script: src/data/run_simulation.py
sandbox: false
token_name: ZENODO_TOKEN
title: Random numbers
description: >-
This is a collection of ten datasets, each containing
ten iid zero-mean, unit-variance random numbers. These
are used in an example of the showyourwork open source
scientific article workflow.
creators:
- Luger, Rodrigo
contents:
*results
The first time we listed all our results files, we added an anchor (&results
),
which we then refer to as an alias (*results
) the next time we need to
list those files. The anchor/alias syntax can help make YAML files shorter
and more readable.
Dependency tarballs (advanced)¶
Sometimes our procedure for generating the results of a simulation might be more involved
than simply executing a single python script. Consider the following example: we have
a file src/data/run_simulation.py
that accepts a number as input, runs a simulation,
and then generates a dataset corresponding to that unique input.
We would then like to ingest all of these datasets into a plotting script that generates
a figure in the paper, all while taking advantage of the Zenodo tarball functionality
in showyourwork
.
Achieving this is simple: let’s re-use the showyourwork.yml
file from the previous
example, but this time omit the script
key:
dependencies:
src/figures/my_figure.py: &results
- src/data/results_0.dat
- src/data/results_1.dat
- src/data/results_2.dat
- src/data/results_3.dat
- src/data/results_4.dat
- src/data/results_5.dat
- src/data/results_6.dat
- src/data/results_7.dat
- src/data/results_8.dat
- src/data/results_9.dat
zenodo:
- src/data/results.tar.gz:
sandbox: false
token_name: ZENODO_TOKEN
title: Random numbers
description: >-
This is a collection of ten datasets, each containing
ten iid zero-mean, unit-variance random numbers. These
are used in an example of the showyourwork open source
scientific article workflow.
creators:
- Luger, Rodrigo
contents:
*results
Instead of the script
key, we include a custom rule
in the Snakefile
to generate all of the results files:
if not config["CI"]:
rule analysis:
input:
"src/data/run_simulation.py"
output:
"src/data/results_{value}.dat"
shell:
"python {input[0]} {wildcards.value}"
That’s it! showyourwork
automatically infers that it must execute the
analysis
rule with all values of the (integer) wildcard value
in the range [0, 10)
to produce the dependencies of the figure script.
When run locally, it will tar them up and upload the tarball to Zenodo;
when running on GitHub Actions, it will download and unpack the tarball
instead of running the analysis
rule.
Note that we explicitly placed the analysis
rule inside a branch that gets
executed only if config["CI"] is False
. The config["CI"]
variable
is automatically set in all showyourwork
workflows: it’s always
False
, except when the workflow is being executed on a GitHub Actions
runner.
Many, many dependencies¶
Sometimes, it might be a pain to explicitly list all the dependencies for a script. This could
be the case if your simulation or analysis step produces many (tens, hundreds, or even thousands of)
data files. The recommended way of dealing with this is through jinja
templating.
For example, instead of explicitly listing all the dependencies in showyourwork.yml
:
showyourwork.yml:
dependencies:
src/figures/my_figure.py:
- src/data/results_0.dat
- src/data/results_1.dat
- src/data/results_2.dat
- src/data/results_3.dat
- src/data/results_4.dat
- src/data/results_5.dat
- src/data/results_6.dat
- src/data/results_7.dat
- src/data/results_8.dat
- src/data/results_9.dat
we recommend deleting the config file showyourwork.yml
and creating
a new file called showyourwork.yml.jinja
with the following template:
showyourwork.yml.jinja:
dependencies:
src/figures/my_figure.py:
{% for i in range(10) -%}
- src/data/results_{{i}}.dat
{% endfor %}
If you’re not familiar with jinja
templating, check out the
documentation. The idea here
is to define a for loop over the variable i
to list all the dependencies
for us. But since this file isn’t a valid YAML config file, we have to add
a bit of boilerplate at the very top of our Snakefile
:
Snakefile:
# Render the config file
import jinja2
with open("showyourwork.yml", "w") as f:
env = jinja2.Environment(loader=jinja2.FileSystemLoader("."))
print(env.get_template("showyourwork.yml.jinja").render(), file=f)
# User config
configfile: "showyourwork.yml"
This code loads the jinja
template, parses it, and prints it to the
actual config file showyourwork.yml
that is ingested by Snakemake
.
This way, showyourwork
only ever sees the config file with all of the
dependencies listed out explicitly, without any of the headaches associated
with listing them all yourself.
Note that if you take this approach, it’s a good idea to add
showyourwork.yml
to your top-level .gitignore
file so that it’s never committed (since
this file is programmatically generated every time you run your workflow).
Custom figure scripts¶
showyourwork
allows you to specify custom scripts for figures. This is useful when showyourwork
can’t
automatically determine the figure script, such as when a figure is
included outside of a figure
environment. The easiest way is to subclass the
figure
rule defined in the showyourwork
module:
# Subclass the `figure` rule to specify that the figure
# `src/figures/custom_figure.pdf` is generated from the script
# `src/figures/custom_script.py`
use rule figure from showyourwork as custom_figure with:
input:
"src/figures/custom_script.py",
"environment.yml"
output:
"src/figures/custom_figure.pdf"
Alternatively, you may override the internal figure
rule completely:
rule custom_figure:
input:
"src/figures/custom_script.py",
"environment.yml",
output:
"src/figures/custom_figure.pdf"
conda:
"environment.yml"
shell:
"cd src/figures && python custom_script.py"
This can be used to execute arbitrary commands for generating figures, such
as producing a figure via a language other than Python, or producing a figure
from a Jupyter notebook. Note that in both cases, showyourwork
expects that
the first file listed under input
is the main script associated with the
figure, and this is what the link in the figure caption will point to on GitHub.
Figures that require LaTeX¶
If you set matplotlib.rc("text", usetex=True)
in your Python
script, you’ll likely
get an error on GitHub Actions complaining that it can’t find latex
. That’s because the
engine used to compile your TeX article – tectonic
– is not a standard TeX
distribution. We recommend disabling the usetex
option in matplotlib
, since
the most common math-mode commands can be rendered using the built-in mathtext
;
see the matplotlib docs.
If, however, you really do need a TeX installation, you can request it in the
.github/workflows/showyourwork.yml
file as follows:
- name: Build the article PDF
id: build
uses: ./showyourwork/showyourwork-action
with:
install-tex: true
env:
ZENODO_TOKEN: ${{ secrets.ZENODO_TOKEN }}
This will install TinyTex, a very lightweight
TeX distribution, on the GitHub Actions runner. Note that TeX rendering in matplotlib
requires certain packages. By default, showyourwork
installs type1cm
and
cm-super
. If you get an error message saying a package is not found, you can request
a package called <package>
to be installed as follows:
- name: Build the article PDF
id: build
uses: ./showyourwork/showyourwork-action
with:
install-tex: true
tex-packages: |
type1cm
cm-super
<package>
env:
ZENODO_TOKEN: ${{ secrets.ZENODO_TOKEN }}
Using graphicspath¶
Users can take advantage of the graphicspath
LaTeX command to specify a
path for all the figures in the workflow. The following snippet
\graphicspath{{./figures/}}
\begin{figure}[ht!]
\begin{centering}
\includegraphics[width=0.4\linewidth]{foo.pdf}
\label{fig:foo}
\end{centering}
\end{figure}
is therefore equivalent to
\begin{figure}[ht!]
\begin{centering}
\includegraphics[width=0.4\linewidth]{figures/foo.pdf}
\label{fig:foo}
\end{centering}
\end{figure}
Note that only a single graphicspath
call is supported, with only
a single path. Additional calls / paths will be ignored by showyourwork
and can lead to errors.
Other LaTeX classes¶
We are slowly adding support for LaTeX classes other than AASTeX. If you don’t see what you’re looking for here, please open an issue.
MNRAS | |||
A&A |
Custom manuscript name¶
By default, showyourwork
expects the manuscript to be called src/ms.tex
. This
can be customized; for example, if you wish to call your article src/article.tex
,
specify it in the showyourwork.yml
file:
ms:
src/article.tex
Note that showyourwork
still expects it to live in the src
directory.
Custom manuscript dependencies¶
Sometimes it is useful to specify custom dependencies of the manuscript file, such as when
you use the input
directive to ingest content from other LaTeX files. Say we have a special
rules that computes the answer to some problem in the Snakefile
:
rule answer:
input:
"src/answer.py"
output:
"src/answer.tex"
conda:
"environment.yml"
shell:
"cd src && python answer.py"
In order to tell showyourwork
that src/ms.tex
depends on src/answer.tex
, simply
specify it as a dependency in showyourwork.yml
:
dependencies:
src/ms.tex:
- src/answer.tex
Because of the way Snakemake
builds the dependency graph, changes to src/answer.py
will automatically trigger re-computation of src/answer.tex
and a rebuild of the PDF.
Non-Python figure scripts¶
Although showyourwork
expects figures to be generated from Python
scripts, it allows
users to provide instructions on how to generate figures using any programming language.
This is done in the showyourwork.yml
config file under the scripts
key. Each entry
should be a file extension, such as sh
for shell scripts, jl
for Julia
scripts,
etc. Under each extension key, users should provide the shell command for generating a figure
from the corresponding script.
For example, the default configuration for Python
scripts looks like this:
scripts:
py:
python {script}
This tells showyourwork
that to generate a figure from a Python
script, all it
needs to do is run the python
shell command followed by the script name, which we
provide as {script}
(this special variable gets automatically expanded at runtime to
the name–not the path–of the script file).
Users don’t need to specify this, however, as Python
is the default language.
The example linked to above shows a more realistic use case: generating a directed
acyclic graph (DAG) from a Graphviz .gv
file:
scripts:
gv:
dot -Tpdf {script} > {figure}
Here, we tell showyourwork
that we need to run the dot
command to generate
the figure {figure}
from the Graphviz script {script}
. Note that {figure}
is another special variable that gets expanded to the name of the output figure file.
Note that because showyourwork
expects scripts to have the .py
extension
by default, you might have to force-add (i.e., git add -f script.gv
) scripts
with other extensions in order to actually commit them!