Using Matplotlib

Matplotlib is a Python plotting library which produces a large variety of visuals. There are several ways you can use Matplotlib in DSS:

  • To display charts within a Jupyter notebook

  • To display charts on a Dashboard

Documentation for Matplotlib is available at https://matplotlib.org/

Installing Matplotlib

  • If you are using the DSS built-in environment, Matplotlib is already installed. You don’t need to do any specific installation

  • If you are using a code environment, you need to install the matplotlib package

Displaying charts in a Jupyter notebook

To display Matplotlib charts in a Jupyter notebook, the easiest is to simply execute the first cell of the notebook:

%pylab inline

This automatically imports the matplotlib and matplotlib.pyplot as plt packages. It also imports numpy as np.

You can then use the regular Matplotlib functions.

For example, to reproduce the simplest sample “sin” figure, use:

# Data for plotting
t = np.arange(0.0, 2.0, 0.01)
s = np.sin(2 * np.pi * t)

plt.plot(t, s)

The chart displays inline in the Jupyter notebook

Displaying Matplotlib charts on a dashboard

Matplotlib charts generated using Python code can be shared on a DSS dashboard using the “static insights” system. -

Each Matplotlib figure can become a single insight in the dashboard.

To do so, create static insights

from dataiku import insights

# This form saves the last displayed figure
insights.save_figure("my-matplotlib-plot")

# If f is a matplotlib figure object, you can save explicitly this figure
# rather than the last displayed one

insights.save_figure("my-matplotlib-explicit-plot", f)

From the Dashboard, you can then add a new “Static” insight, select the my-matplotlib-plot insight

Refreshing charts on a dashboard

You can refresh the charts automatically on a dashboard by using a scenario to re-run the above piece of code.

This call to dataiku.insights code can be:

  • In a DSS recipe (use a regular “Build” scenario step)

  • In a Jupyter notebook (use a “Export notebook” scenario step)

  • As a custom Python scenario step

Using matplotlib in a recipe

Matplotlib is based on a concept of backends. Each backend knows how to display figures. When running Python on your local machine, Matplotlib will by default pop a graphical window (a Windows, macOS or Linux window) to show each plot. This is done using a specific backend for each OS.

In a notebook, when using `%pylab inline`, it automatically activates a specific backend that displays the plots inline within the Jupyter interface.

In a recipe or in a custom Python scenario step, the notebook backend is not available. You must thus, as the very first action in your recipe, use the following code:

import matplotlib
matplotlib.use("Agg")

This forces the use of the “Agg” backend which performs rendering of the charts in-memory. Charts can then be exported to PNG, …

Warning

This matplotlib.use("Agg") must take place before import matplotlib.pyplot. Failure to do so will generally result in an error with “Tkinter”. This error is caused by the fact that Matplotlib will by default try to import a backend called “TCL/TK” which is based on the “Tkinter” Python library, which is not usually installed