jupytext 1.0.1rc0

Jupyter notebooks as Markdown documents, Julia, Python or R scripts

Jupyter notebooks as Markdown documents, Julia, Python or R scripts

Have you always wished Jupyter notebooks were plain text documents? Wished you could edit them in your favorite IDE? And get clear and meaningfull diffs when doing version control? Then... Jupytext may well be the tool you're looking for!

Jupytext can save Jupyter notebooks as

• Markdown and R Markdown documents,
• Julia, Python, R, Bash, Scheme, C++ and q/kdb+ scripts.

There are multiple ways to use jupytext:

• Directly from Jupyter Notebook or JupyterLab. Jupytext provides a contents manager that allows Jupyter to save your notebook to your favorite format (.py, .R, .jl, .md, .Rmd...) in addition to (or in place of) the traditional .ipynb file. The text representation can be edited in your favorite editor. When you're done, refresh the notebook in Jupyter: inputs cells are loaded from the text file, while output cells are reloaded from the .ipynb file if present. Refreshing preserves kernel variables, so you can resume your work in the notebook and run the modified cells without having to rerun the notebook in full.
• On the command line. jupytext converts Jupyter notebooks to their text representation, and back. The command line tool can act on noteboks in many ways. It can synchronize multiple representations of a notebook, pipe a notebook into a reformatting tool like black, etc... It can also work as a pre-commit hook if you wish to automatically update the text representation when you commit the .ipynb file.
• in Vim: edit your Jupyter notebooks, represented as a Markdown document, or a Python script, with jupytext.vim.

Demo time

Looking for a demo?

• Read the original announcement in Towards Data Science,
• Watch the PyParis talk,
• or, try Jupytext online with binder!

Example usage

Writing notebooks as plain text

You like to work with scripts? The good news is that plain scripts, which you can draft and test in your favorite IDE, open transparently as notebooks in Jupyter when using Jupytext. Run the notebook in Jupyter to generate the outputs, associate an .ipynb representation, save and share your research as either a plain script or as a traditional Jupyter notebook with outputs.

Collaborating on Jupyter Notebooks

With Jupytext, collaborating on Jupyter notebooks with Git becomes as easy as collaborating on text files.

The setup is straightforward:

• Open your favorite notebook in Jupyter notebook
• Associate a .py representation (for instance) to that notebook
• Save the notebook, and put the Python script under Git control. Sharing the .ipynb file is possible, but not required.

Collaborating then works as follows:

• Your collaborator pulls your script.
• The script opens as a notebook in Jupyter, with no outputs (in JupyterLab this requires a right-click).
• They run the notebook and save it. Outputs are regenerated, and a local .ipynb file is created.
• Note that, alternatively, the .ipynb file could have been regenerated with jupytext --sync notebook.py.
• They change the notebook, and push their updated script. The diff is nothing else than a standard diff on a Python script.
• You pull the changed script, and refresh your browser. Input cells are updated. The outputs from cells that were changed are removed. Your variables are untouched, so you have the option to run only the modified cells to get the new outputs.

Code refactoring

In the animation below we propose a quick demo of Jupytext. While the example remains simple, it shows how your favorite text editor or IDE can be used to edit your Jupyter notebooks. IDEs are more convenient than Jupyter for navigating through code, editing and executing cells or fractions of cells, and debugging.

• We start with a Jupyter notebook.
• The notebook includes a plot of the world population. The plot legend is not in order of decreasing population, we'll fix this.
• We want the notebook to be saved as both a .ipynb and a .py file: we select Pair Notebook with a light Script in the File/Jupytext menu, which has the effet to add a "jupytext": {"formats": "ipynb,py:light"}, entry to the notebook metadata.
• The Python script can be opened with PyCharm:
  • Navigating in the code and documentation is easier than in Jupyter.
  • The console is convenient for quick tests. We don't need to create cells for this.
  • We find out that the columns of the data frame were not in the correct order. We update the corresponding cell, and get the correct plot.
• The Jupyter notebook is refreshed in the browser. Modified inputs are loaded from the Python script. Outputs and variables are preserved. We finally rerun the code and get the correct plot.

Importing Jupyter Notebooks as modules

Jupytext allows to import code from other Jupyter notebooks in a very simple manner. Indeed, all you need to do is to pair the notebook that you wish to import with a script, and import the resulting script.

If the notebook contains demos and plots that you don't want to import, mark those cell as either

• active only in the ipynb format, with the {"active": "ipynb"} cell metadata
• frozen, using the freeze extension for Jupyter notebook.

Inactive cells will be commented in the paired script, and consequently will not be executed when the script is imported.

Installation

Jupytext is available on pypi and on conda-forge. Run either of

pip install jupytext --upgrade

or

conda install -c conda-forge jupytext

If you want to use Jupytext within Jupyter Notebook or JupyterLab, make sure you install Jupytext in the Python environment where the Jupyter server runs. If that environment is read-only, for instance if your server is started using JupyterHub, install Jupytext in user mode with:

/path_to_your_jupyter_environment/python -m pip install jupytext --upgrade --user

Jupytext's content manager

Jupytext includes a contents manager for Jupyter that allows Jupyter to open and save notebooks as text files. When Jupytext's content manager is active in Jupyter, scripts and Markdown documents have a notebook icon.

If you don't have the notebook icon on text documents after a fresh restart of your Jupyter server, install the contents manager manually. Append

c.NotebookApp.contents_manager_class = "jupytext.TextFileContentsManager"

to your .jupyter/jupyter_notebook_config.py file (generate a Jupyter config, if you don't have one yet, with jupyter notebook --generate-config). Our contents manager accepts a few options: default formats, default metadata filter, etc — read more on this below. Then, restart Jupyter Notebook or JupyterLab, either from the JupyterHub interface or from the command line with

jupyter notebook # or lab

Jupytext menu in Jupyter Notebook

Jupytext includes an extensions for Jupyter Notebook that adds a Jupytext section in the File menu.

If the extension was not automatically installed, install and activate it with

jupyter nbextension install --py jupytext [--user]
jupyter nbextension enable --py jupytext [--user]

Jupytext commands in JupyterLab

In JupyterLab, Jupytext adds a set of commands to the command palette:

If you don't see these commands, install the extension manually with

jupyter labextension install jupyterlab-jupytext

(the above requires npm, run conda install nodejs first if you don't have npm).

Paired notebooks

Jupytext can write a given notebook to multiple files. In addition to the original notebook file, Jupytext can save the input cells to a text file — either a script or a Markdown document. Put the text file under version control for a clear commit history. Or refactor the paired script, and reimport the updated input cells by simply refreshing the notebook in Jupyter.

Per-notebook configuration

Select the pairing for a given notebook using either the Jupytext menu in Jupyter Notebook, or the Jupytext commands in JupyterLab.

These command simply add a "jupytext": {"formats": "ipynb,md"}-like entry in the notebook metadata. You could also set that metadata yourself with Edit/Edit Notebook Metadata in Jupyter Notebook. In JupyterLab, use this extension.

The pairing information for one or multiple notebooks can be set on the command line:

jupytext --set-formats ipynb,py [--sync] notebook.ipynb

You can pair a notebook to as many text representations as you want (see our World population notebook in the demo folder). Format specifications are of the form

[[path/][prefix]/][suffix.]ext[:format_name]

where

• ext is one of ipynb, md, Rmd, jl, py, R, sh, cpp, q. Use the auto extension to have the script extension chosen according to the Jupyter kernel.
• format_name (optional) is either light (default for scripts), bare, percent, hydrogen, sphinx (Python only), spin (R only) — see below for the format specifications.
• path, prefix and suffix allow to save the text representation to files with different names, or in a different folder. For instance, if you want that your notebook is paired to a python script in a subfolder named scripts, set the formats metadata to ipynb,scripts//py. If the notebook is in a notebooks folder and you want the text representation to be in a scripts folder at the same level, use notebooks//ipynb,scripts//py.

Jupytext accepts a few additional options. These options should be added to the "jupytext" section in the metadata — use either the metadata editor or the --opt/--format-options argument on the command line.

• comment_magics: By default, Jupyter magics are commented when notebooks are exported to any other format than markdown. If you prefer otherwise, use this boolean option, or is global counterpart (see below).
• notebook_metadata_filter: By default, Jupytext only exports the kernelspec and jupytext metadata to the text files. Set "jupytext": {"notebook_metadata_filter": "-all"} if you want that the script has no notebook metadata at all. The value for notebook_metadata_filter is a comma separated list of additional/excluded (negated) entries, with all a keyword that allows to exclude all entries.
• cell_metadata_filter: By default, cell metadata autoscroll, collapsed, scrolled, trusted and ExecuteTime are not included in the text representation. Add or exclude more cell metadata with this option.

Global configuration

Jupytext's contents manager also accepts global options. We start with the default format pairing. Say you want to always associate every .ipynb notebook with a .md file (and reciprocally). This is simply done by adding the following to your Jupyter configuration file:

# Always pair ipynb notebooks to md files
c.ContentsManager.default_jupytext_formats = "ipynb,md"

(and similarly for the other formats).

In case the percent format is your favorite, add the following to your .jupyter/jupyter_notebook_config.py file:

# Use the percent format when saving as py
c.ContentsManager.preferred_jupytext_formats_save = "py:percent"

and then, Jupytext will understand "jupytext": {"formats": "ipynb,py"} as an instruction to create the paired Python script in the percent format.

To disable global pairing for an individual notebook, set formats to a single format, e.g.: "jupytext": {"formats": "ipynb"}

Metadata filtering

You can specify which metadata to include or exclude in the text files created by Jupytext by setting c.ContentsManager.default_notebook_metadata_filter (notebook metadata) and c.ContentsManager.default_cell_metadata_filter (cell metadata). They accept a string of comma separated keywords. A minus sign - in font of a keyword means exclusion.

Suppose you want to keep all the notebook metadata but widgets and varInspector in the YAML header. For cell metadata, you want to allow ExecuteTime and autoscroll, but not hide_output. You can set

c.ContentsManager.default_notebook_metadata_filter = "all,-widgets,-varInspector"
c.ContentsManager.default_cell_metadata_filter = "ExecuteTime,autoscroll,-hide_output"

If you want that the text files created by Jupytext have no metadata, you may use the global metadata filters below. Please note that with this setting, the metadata is only preserved in the .ipynb file.

c.ContentsManager.default_notebook_metadata_filter = "-all"
c.ContentsManager.default_cell_metadata_filter = "-all"

NB: All these global options (and more) are documented here.

How to edit the notebook simultaneously in Jupyter and a text editor?

When editing a paired notebook using Jupytext's contents manager, Jupyter updates both the .ipynb and its text representation. The text representation can be edited outside of Jupyter. When the notebook is refreshed in Jupyter, the input cells are read from the text file, and the output cells from the .ipynb file.

It is possible (and convenient) to leave the notebook open in Jupyter while you edit its text representation. However, you don't want that the two editors save the notebook simultaneously. To avoid this:

• deactivate Jupyter's autosave, by toggling the "Autosave notebook" menu entry (or run %autosave 0 in a cell of the notebook)
• and refresh the notebook when you switch back from the editor to Jupyter.

In case you forgot to refresh, and saved the Jupyter notebook while the text representation had changed, no worries: Jupyter will ask you which version you want to keep:

When that occurs, please choose the version in which you made the latest changes. And give a second look to our advice to deactivate the autosaving of notebooks in Jupyter.

Command line conversion

The package provides a jupytext script for command line conversion between the various notebook extensions:

jupytext --to python notebook.ipynb             # create a notebook.py file
jupytext --to py:percent notebook.ipynb         # create a notebook.py file in the double percent format
jupytext --to py:percent --comment-magics false notebook.ipynb   # create a notebook.py file in the double percent format, and do not comment magic commands
jupytext --to markdown notebook.ipynb           # create a notebook.md file
jupytext --output script.py notebook.ipynb      # create a script.py file

jupytext --to notebook notebook.py              # overwrite notebook.ipynb (remove outputs)
jupytext --to notebook --update notebook.py     # update notebook.ipynb (preserve outputs)
jupytext --to ipynb notebook1.md notebook2.py   # overwrite notebook1.ipynb and notebook2.ipynb

jupytext --to md --test notebook.ipynb          # Test round trip conversion

jupytext --to md --output - notebook.ipynb      # display the markdown version on screen
jupytext --from ipynb --to py:percent           # read ipynb from stdin and write double percent script on stdout

Jupytext has a --sync mode that updates all the paired representations of a notebook based on the file that was last modified. You may also find useful to --pipe the text representation of a notebook into tools like black:

jupytext --sync --pipe black notebook.ipynb    # read most recent version of notebook, reformat with black, save

The jupytext command accepts many arguments. Use the --set-formats and the --update-metadata arguments to edit the pairing information or more generally the notebook metadata. Execute jupytext --help to access the documentation.

Jupytext as a Git pre-commit hook

Jupytext is also available as a Git pre-commit hook. Use this if you want Jupytext to create and update the .py (or .md...) representation of the staged .ipynb notebooks. All you need is to create an executable .git/hooks/pre-commit file with the following content:

#!/bin/sh
# For every ipynb file in the git index, add a Python representation
jupytext --from ipynb --to py:light --pre-commit

#!/bin/sh
# For every ipynb file in the git index:
# - apply black and flake8
# - export the notebook to a Python script in folder 'python'
# - and add it to the git index
jupytext --from ipynb --pipe black --check flake8 --pre-commit
jupytext --from ipynb --to python//py:light --pre-commit

If you don't want notebooks to be committed (and only commit the representations), you can ask the pre-commit hook to unstage notebooks after conversion by adding the following line:

git reset HEAD **/*.ipynb

Note that these hooks do not update the .ipynb notebook when you pull. Make sure to either run jupytext in the other direction, or to use our paired notebook and our contents manager for Jupyter. Also, Jupytext does not offer a merge driver. If a conflict occurs, solve it on the text representation and then update or recreate the .ipynb notebook. Or give a try to nbdime and its merge driver.

Reading notebooks in Python

Manipulate notebooks in a Python shell or script using jupytext's main functions:

# Read a notebook from a file. Format can be any of 'py', 'md', 'jl:percent', ...
readf(nb_file, fmt=None)

# Read a notebook from a string. Here, format should contain at least the file extension.
reads(text, fmt)

# Return the text representation for a notebook in the desired format.
writes(notebook, fmt)

# Write a notebook to a file in the desired format.
writef(notebook, nb_file, fmt=None)

Round-trip conversion

Representing Jupyter notebooks as scripts requires a solid round trip conversion. You don't want your notebooks (nor your scripts) to be modified because you are converting them to the other form. A few hundred tests ensure that round trip conversion is safe.

You can easily test that the round trip conversion preserves your Jupyter notebooks and scripts. Run for instance:

# Test the ipynb -> py:percent -> ipynb round trip conversion
jupytext --test notebook.ipynb --to py:percent

# Test the ipynb -> (py:percent + ipynb) -> ipynb (à la paired notebook) conversion
jupytext --test --update notebook.ipynb --to py:percent

Note that jupytext --test compares the resulting notebooks according to its expectations. If you wish to proceed to a strict comparison of the two notebooks, use jupytext --test-strict, and use the flag -x to report with more details on the first difference, if any.

Please note that

• Scripts opened with Jupyter have a default metadata filter that prevents additional notebook or cell metadata to be added back to the script. Remove the filter if you want to store Jupytext's settings, or the kernel information, in the text file.
• Cell metadata are available in light and percent formats for all cell types. Sphinx Gallery scripts in sphinx format do not support cell metadata. R Markdown and R scripts in spin format support cell metadata for code cells only. Markdown documents do not support cell metadata.
• By default, a few cell metadata are not included in the text representation of the notebook. And only the most standard notebook metadata are exported. Learn more on this in the sections for notebook specific and global settings for metadata filtering.
• Representing a Jupyter notebook as a Markdown or R Markdown document has the effect of splitting markdown cells with two consecutive blank lines into multiple cells (as the two blank line pattern is used to separate cells).

Format specifications

Markdown and R Markdown

Save Jupyter notebooks as Markdown documents and edit them in one of the many editors with good Markdown support.

R Markdown is RStudio's format for notebooks, with support for R, Python, and many other languages.

Our implementation for Jupyter notebooks as Markdown or R Markdown documents is straightforward:

• A YAML header contains the notebook metadata (Jupyter kernel, etc)
• Markdown cells are inserted verbatim, and separated with two blank lines
• Code and raw cells start with triple backticks collated with cell language, and end with triple backticks. Cell metadata are not available in the Markdown format. The code cell options in the R Markdown format are mapped to the corresponding Jupyter cell metadata options, when available.

See how our World population.ipynb notebook in the demo folder is represented in Markdown or R Markdown.

When editing Jupyter Markdown, you can split text into markdown cells by adding two blank lines at the point you want the text to split. This is the default rule, but you may want to modify the rule for the case of Markdown headers in text. By default, a single blank line followed by a Markdown header will not cause the cell to split, so the header will appear in the middle of a text cell. You may prefer to always split text cells at headers. If so, use the split_at_heading option. Set the option either on the command line, or by adding "split_at_heading": true to the jupytext section in the notebook metadata, or on Jupytext's content manager:

c.ContentsManager.split_at_heading = True

This will cause jupytext to split markdown text cells at heading prefixed by one blank line, so the heading appears at the top of a new cell. Without this option, you would need two blank lines above the heading to cause the split.

The light format for notebooks as scripts

The light format was created for this project. It is the default format for scripts. That format can read any script as a Jupyter notebook, even scripts which were never prepared to become a notebook. When a notebook is written as a script using this format, only a few cells markers are introduced—none if possible.

The light format has:

• A (commented) YAML header, that contains the notebook metadata.
• Markdown cells are commented, and separated with a blank line.
• Code cells are exported verbatim (except for Jupyter magics, which are commented), and separated with blank lines. Code cells are reconstructed from consistent Python paragraphs (no function, class or multiline comment will be broken).
• Cells that contain more than one Python paragraphs need an explicit start-of-cell delimiter # + (// + in C++, etc). Cells that have explicit metadata have a cell header # + {JSON} where the metadata is represented, in JSON format. The end of cell delimiter is # -, and is omitted when followed by another explicit start of cell marker.

The light format is currently available for Python, Julia, R, Bash, Scheme, C++ and q/kdb+. Open our sample notebook in the light format here.

A variation of the light format is the bare format, with no cell marker at all. Please note that this format will split your code cells on code paragraphs. By default, this format still includes a YAML header - if you prefer to also remove the header, set "notebook_metadata_filter": "-all" in the jupytext section of your notebook metadata.

The percent format

The percent format is a representation of Jupyter notebooks as scripts, in which cells are delimited with a commented double percent sign # %%. The format was introduced by Spyder five years ago, and is now supported by many editors, including

• Spyder IDE,
• Hydrogen, a package for Atom,
• VS Code with the vscodeJupyter extension,
• Python Tools for Visual Studio,
• and PyCharm Professional.

Our implementation of the percent format is compatible with the original specifications by Spyder. We extended the format to allow markdown cells and cell metadata. Cell headers have the following structure:

# %% Optional text [cell type] {optional JSON metadata}

where cell type is either omitted (code cells), or [markdown] or [raw]. The content of markdown and raw cells is commented out in the resulting script.

Percent scripts created by Jupytext have a header with an explicit format information. The format of scripts with no header is inferred automatically: scripts with at least one # %% cell are identified as percent scripts. Scripts with at least one double percent cell, and an uncommented Jupyter magic command, are identified as hydrogen scripts.

The percent format is currently available for Python, Julia, R, Bash, Scheme, C++ and q/kdb+. Open our sample notebook in the percent format here.

If the percent format is your favorite one, add the following to your .jupyter/jupyter_notebook_config.py file:

c.ContentsManager.preferred_jupytext_formats_save = "py:percent" # or "auto:percent"

Then, Jupytext's content manager will understand "jupytext": {"formats": "ipynb,py"}, as an instruction to create the paired Python script in the percent format.

By default, Jupyter magics are commented in the percent representation. If you run the percent scripts in Hydrogen, use instead the hydrogen format, a variant of the percent format that does not comment Jupyter magic commands.

Sphinx-gallery scripts

Another popular notebook-like format for Python scripts is the Sphinx-gallery format. Scripts that contain at least two lines with more than twenty hash signs are classified as Sphinx-Gallery notebooks by Jupytext.

Comments in Sphinx-Gallery scripts are formatted using reStructuredText rather than markdown. They can be converted to markdown for a nicer display in Jupyter by adding a c.ContentsManager.sphinx_convert_rst2md = True line to your Jupyter configuration file. Please note that this is a non-reversible transformation—use this only with Binder. Revert to the default value sphinx_convert_rst2md = False when you edit Sphinx-Gallery files with Jupytext.

Turn a GitHub repository containing Sphinx-Gallery scripts into a live notebook repository with Binder and Jupytext by adding only two files to the repo:

• binder/requirements.txt, a list of the required packages (including jupytext)
• .jupyter/jupyter_notebook_config.py with the following contents:

c.NotebookApp.contents_manager_class = "jupytext.TextFileContentsManager"
c.ContentsManager.preferred_jupytext_formats_read = "py:sphinx"
c.ContentsManager.sphinx_convert_rst2md = True

Our sample notebook is also represented in sphinx format here.

R knitr::spin scripts

The spin format is specific to R scripts. These scripts can be compiled into reports using either knitr::spin or RStudio. The implementation of the format is as follows:

• Jupyter metadata are in YAML format, in a #' -commented header.
• Markdown cells are commented with #' .
• Code cells are exported verbatim. Cell metadata are signalled with #+. Cells end with a blank line, an explicit start of cell marker, or a markdown cell.

Jupyter Notebook or JupyterLab?

Jupytext works well in both. Just note that:

• A notebook metadata editor for JupyterLab is available here.
• JupyterLab can open any paired notebook with extension .ipynb. Paired notebooks work exactly as in Jupyter Notebook: input cells are taken from the text notebook, and outputs from the .ipynb file. Both files are updated when the notebook is saved.
• In JupyterLab, scripts or Markdown documents open as text by default. Open them as notebooks with the Open With -> Notebook context menu (available in JupyterLab 0.35 and above) as shown below. If the text document is a paired notebook, then the associated .ipynb file will be regenerated when you save the document (alternatively, you could have recreated the missing .ipynb file with jupytext --sync).

Fine tuning

Jupyter magic commands are commented when exporting the notebook to text, except for the markdown and the hydrogen format. If you want to change this for a single line, add a #escape or #noescape flag on the same line as the magic, or a "comment_magics": true or false entry in the notebook metadata, in the "jupytext" section. Or set your preference globally on the contents manager by adding this line to .jupyter/jupyter_notebook_config.py:

c.ContentsManager.comment_magics = True # or False

Also, you may want some cells to be active only in the Python, or R Markdown representation. For this, use the active cell metadata. Set "active": "ipynb" if you want that cell to be active only in the Jupyter notebook. And "active": "py" if you want it to be active only in the Python script. And "active": "ipynb,py" if you want it to be active in both, but not in the R Markdown representation...

Extending the light and percent formats to more languages

You want to extend the light and percent format to another language? In principle that is easy, and you will only have to:

• document the language extension and comment by adding one line to _SCRIPT_EXTENSIONS in languages.py.
• contribute a sample notebook in tests\notebooks\ipynb_[language].
• add two tests in test_mirror.py: one for the light format, and another one for the percent format.
• Make sure that the tests pass, and that the text representations of your notebook, found in tests\notebooks\mirror\ipynb_to_script and tests\notebooks\mirror\ipynb_to_percent, are valid scripts.

Want to contribute?

Contributions are welcome. Please let us know how you use jupytext and how we could improve it. You think the documentation could be improved? Go ahead! And stay tuned for more demos on medium and twitter!