Kedro and Jupyter Notebooks

You can take advantage of a Jupyter Notebook’s liberal development environment for exploratory data analysis and experimentation from within a Kedro project. Later, when you need to follow software best practices as the project complexity increases, or as you scale into production, you can transfer code from the Notebook into Kedro to benefit from its opinionated project framework.

This page explains how best to combine Kedro and Jupyter Notebook development and illustrates with an example Notebook that has access to the catalog, context, pipelines and session variables for a Kedro project.

A custom Kedro kernel

Kedro offers a command (kedro jupyter notebook) to create a Jupyter kernel named kedro_<package_name> that is almost identical to the default IPython kernel but with a slightly customised kernel specification.

The custom kernel automatically loads kedro.ipython, which is an IPython extension that launches a Kedro session and makes the following Kedro variables available:

  • catalog (type DataCatalog): Data Catalog instance that contains all defined datasets; this is a shortcut for context.catalog

  • context (type KedroContext): Kedro project context that provides access to Kedro’s library components

  • pipelines (type Dict[str, Pipeline]): Pipelines defined in your pipeline registry

  • session (type KedroSession): Kedro session that orchestrates a pipeline run

Iris dataset example

Create a sample Kedro project with the pandas-iris starter as we showed in the get started documentation:

kedro new --starter=pandas-iris

We will assume you call the project iris, but you can call it whatever you choose.

Navigate to the project directory and issue the following command in the terminal to launch Jupyter:

kedro jupyter notebook

Your browser window will open, and you can then create a new Notebook using the dropdown and selecting the Kedro (<project_name>) kernel.

Create a new Notebook with Kedro (iris) kernel

We recommend that you store your Notebooks in the notebooks folder of your Kedro project.

We will now give some examples of how to work with the Kedro variables. To explore the full range of attributes and methods available, you might like to consult the relevant API documentation or use the Python dir function (e.g. dir(catalog)).


If the Kedro variables are not available within your Notebook, you could have a malformed configuration file or missing dependencies. The full error message is shown on the terminal used to launch kedro jupyter notebook.


catalog can be used to explore your Data Catalog, including parameters. Useful methods include catalog.list, catalog.load and For example, add the following to a cell in your Notebook:


When you run the cell:


Next try the following:


The output:

INFO     Loading data from 'example_iris_data' (CSVDataSet)...

     sepal_length  sepal_width  petal_length  petal_width    species
0             5.1          3.5           1.4          0.2     setosa
1             4.9          3.0           1.4          0.2     setosa
2             4.7          3.2           1.3          0.2     setosa
3             4.6          3.1           1.5          0.2     setosa
4             5.0          3.6           1.4          0.2     setosa
..            ...          ...           ...          ...        ...
145           6.7          3.0           5.2          2.3  virginica
146           6.3          2.5           5.0          1.9  virginica
147           6.5          3.0           5.2          2.0  virginica
148           6.2          3.4           5.4          2.3  virginica
149           5.9          3.0           5.1          1.8  virginica

Finally, try the following:


You should see the following:

INFO     Loading data from 'parameters' (MemoryDataSet)...

{'example_test_data_ratio': 0.2,
 'example_num_train_iter': 10000,
 'example_learning_rate': 0.01}


If you enable versioning you can load a particular version of a dataset, e.g. catalog.load("example_train_x", version="2021-12-13T15.08.09.255Z").


context enables you to access Kedro’s library components and project metadata. For example, if you add the following to a cell and run it:


You should see output similar to the following, according to your username and path:


You can find out more about the context in the API documentation.


pipelines is a dictionary containing your project’s registered pipelines:


The output will be a listing as follows:

{'__default__': Pipeline([
Node(split_data, ['example_iris_data', 'parameters'], ['X_train', 'X_test', 'y_train', 'y_test'], 'split'),
Node(make_predictions, ['X_train', 'X_test', 'y_train'], 'y_pred', 'make_predictions'),
Node(report_accuracy, ['y_pred', 'y_test'], None, 'report_accuracy')

You can use this to explore your pipelines and the nodes they contain:


Should give the output:

{'y_pred', 'X_test', 'y_train', 'X_train', 'y_test'}

session allows you to run a pipeline. With no arguments, this will run your __default__ project pipeline sequentially, much as a call to kedro run from the terminal:


You can only execute one successful run per session, as there’s a one-to-one mapping between a session and a run. If you wish to do multiple runs, you’ll have to run %reload_kedro to obtain a new session (see below).

You can also specify the following optional arguments for

Argument name

Accepted types




Construct the pipeline using only nodes which have this tag attached. A node is included in the resulting pipeline if it contains any of those tags



An instance of Kedro AbstractRunner. Can be an instance of a ParallelRunner



Run only nodes with specified names



A list of node names which should be used as a starting point



A list of node names which should be used as an end point



A list of dataset names which should be used as a starting point



A list of dataset names which should be used as an end point


Dict[str, str]

A mapping of a dataset name to a specific dataset version (timestamp) for loading. Applies to versioned datasets only



Name of the modular pipeline to run. Must be one of those returned by the register_pipelines function in src/<package_name>/

%reload_kedro line magic

You can use %reload_kedro line magic within your Notebook to reload the Kedro variables (for example, if you need to update catalog following changes to your Data Catalog).

You don’t need to restart the kernel to reload the Kedro IPython extension and refresh the catalog, context, pipelines and session variables.

%reload_kedro accepts optional keyword arguments env and params. For example, to use configuration environment prod:

%reload_kedro --env=prod`

For more details, run %reload_kedro?.

%run_viz line magic

If you have Kedro-Viz installed for the project you can display an interactive visualisation of your pipeline directly in your Notebook using the line magic %run_viz.

Convert functions from Jupyter Notebooks into Kedro nodes

If you are writing experimental code in your Notebook and later want to convert functions you’ve written to Kedro nodes, you can do this using tags.

Say you have the following code in your Notebook:

def some_action():
    print("This function came from `notebooks/my_notebook.ipynb`")
  1. Enable tags toolbar: View menu -> Cell Toolbar -> Tags Enable the tags toolbar graphic

  2. Add the node tag to the cell containing your function Add the node tag graphic


The Notebook can contain multiple functions tagged as node, each of them will be exported into the resulting Python file

  1. Save your Jupyter Notebook to notebooks/my_notebook.ipynb

  2. From your terminal, run kedro jupyter convert notebooks/my_notebook.ipynb from the Kedro project directory. The output is a Python file src/<package_name>/nodes/ containing the some_action function definition

  3. The some_action function can now be used in your Kedro pipelines

Useful to know…

Each Kedro project has its own Jupyter kernel so you can switch between multiple Kedro projects from a single Jupyter instance simply by selecting the appropriate kernel.

If a Jupyter kernel with the name kedro_<package_name> already exists then it is replaced. This ensures that the kernel always points to the correct Python executable. For example, if you change conda environment in a Kedro project then you should re-run kedro jupyter notebook to replace the kernel specification with one that points to the new environment.

You can use the jupyter kernelspec set of commands to manage your Jupyter kernels. For example, to remove a kernel, run jupyter kernelspec remove <kernel_name>.

Managed services

If you work within a managed Jupyter service such as a Databricks Notebook you may be unable to execute kedro jupyter notebook. You can explicitly load the Kedro IPython extension with the %load_ext line magic:

In [1]: %load_ext kedro.ipython

If you launch your Jupyter instance from outside your Kedro project, you will need to run a second line magic to set the project path so that Kedro can load the catalog, context, pipelines and session variables:

In [2]: %reload_kedro <project_root>

The Kedro IPython extension remembers the project path so that subsequent calls to %reload_kedro do not need to specify it:

In [1]: %load_ext kedro.ipython
In [2]: %reload_kedro <project_root>
In [3]: %reload_kedro

IPython, JupyterLab and other Jupyter clients

You can also connect an IPython shell to a Kedro project kernel as follows:

kedro ipython

The command launches an IPython shell with the extension already loaded and is equivalent to the command ipython --ext kedro.ipython. You first saw this in action in the spaceflights tutorial.

Similarly, the following creates a custom Jupyter kernel that automatically loads the extension and launches JupyterLab with this kernel selected:

kedro jupyter lab

You can use any other Jupyter client to connect to a Kedro project kernel such as the Qt Console, which can be launched using the kedro_iris kernel as follows:

jupyter qtconsole --kernel=kedro_iris

This will automatically load the Kedro IPython in a console that supports graphical features such as embedded figures: Plot of example iris data in a Qt Console