Data Viz in Fabric Notebooks

Lots of people have created Power BI reports, using interactive data visualizations to explore and communicate data. When Power BI was first created, it was used in situations that weren’t ideal because that was all we had as far as cloud-based tools in the Microsoft data stack. Now, in addition to interactive reports, we have paginated reports and notebooks. In this post, I’ll discuss when notebooks might be an appropriate visualization tool.

High-level comparison

If you haven’t looked into Fabric notebooks for data visualization, the table below offers a high-level comparison to Power BI reports. There are edge cases where you could do something successfully in either tool, but generally certain personas and use cases lean towards one or the other.

Feature	Power BI	Fabric Notebooks
Purpose?	Designed for interactive dashboards and reports?	Used for data exploration, transformation, and advanced analytics?
Visuals?	Drag-and-drop visuals, prebuilt charts?	Code-based visuals using Python (Matplotlib, Seaborn) or Spark libraries?
Interactivity?	Highly interactive, slicers, drill-throughs?	Limited interactivity; static charts unless using specific libraries?
Best For?	Business reporting and storytelling?	Data exploration, debugging, machine learning/AI

Where do notebooks shine?

Notebooks are not usually the best interface for business dashboards and scorecards, but there are other situations where they might be better than a Power BI report:

Data dumps (can save to OneLake or blob storage when output contains a lot of rows)
Detailed static chart images
Maps (interactive or static)
Highly customized charts and uncommon chart types
High-density visuals
Adding advanced statistics, AI, or ML on top of a base dataset before visualization

You can create a table (or tables) of data and output them to an easily accessible location, whether that is downloading the data directly from a notebook or exporting to OneLake. You don’t need to create a visual first, just to export the DataFrame contents.

There are lots of great mapping libraries available for Python, so if you can’t achieve what you need in an Azure Map visual and don’t want to pay for ArcGIS or another embedded service, you may consider creating your map in a notebook. Check out Plotly and DataShader for examples.

A map showing casualties by severity created using Altair — A map created using Altair in a Fabric notebook

Many Power BI visuals have a default of 3,500 data points allowed. Some visuals, such as the Azure Map, allow up to 30,000 data points by default. If you have a situation where the data should not be aggregated or sampled, a python visual may be a good option.

A scatterplot with a dark background and red dots of various levels of opacity. — A scatterplot created using Matplotlib in a Fabric notebook

Sometimes we want to create highly customized visualizations that are just not achievable in the core or free custom Power BI visuals. While you can usually make what you need in Deneb, it may require a lot of effort or you may be introducing a skillset that your team doesn’t have or has chosen not to emphasize. If your team already knows Python, you can explore creating visuals in a notebook and outputting them as SVGs to include in documents, presentations, or websites.

a swarm plot with a gray background and blue dots showing penguin flipper length by spiecies — A swarmplot created using Seaborn in a Fabric notebook

And while it’s possible to add AI or ML to your data using Power Query, it is often easier to use Python libraries to add these capabilities and then visualize the data immediately after.

Interactivity

Python visuals can offer some interactivity, depending on the library. It is often only within one visual. For instance, you might have a drop-down for selecting a category which filters a bar chart. Or you might use a lasso tool to select an area of a map. What is less common is to create a set of multiple charts in a single output that all interact with each other. That is one area where Power BI shines and requires little to no effort, since the default is for visuals to interact with all other visuals on the report page.

Remember your audience

While the visualizations produced by a notebook may be great for your audience, the notebook interface itself might not be ideal for consumption. Data engineers are often used to notebooks in other contexts, but the average business analyst might not want to use a notebook. This may lead to creating the visualization in one tool and consuming it in another. You’ll have to examine the context to and use case to decide if that is appropriate.

Could a notebook really be easier than a drag-and-drop interface?

If writing Python code seems unattainable, a notebook might not be for you. But I’d like to share a couple of thoughts about what has made Python easier for me.

The libraries

After learning about the concept of a nullity matrix, I set out to create one in Power BI for a Workout Wednesday challenge. After experimenting with core visuals and not liking the results, I switched to Deneb. I struggled a bit to get what I needed in Deneb and after an hour or so, I sought out help. The Deneb visual ultimately worked fine, but it was a lot of effort to get there. There is a free library to do this in Python. Once you have a DataFrame with the data you want to include, it’s two lines of code to create the visual.

a nullity matrix that plots column names across the top and rows along the left. There is whitespace when values are null. — A nullity matrix created using the missingno library

AI Coding Assistants

With AI assistants to help us write Python code, the barrier to getting our desired visual output in a notebook is possibly lower than ever. You can integrate AI assistants like GitHub Copilot or Claude Code into VS Code to have a more seamless development experience outside of a browser, if you prefer. Just be sure that your use of AI coding assistants meets any organizational and/or personal requirements around information security and intellectual property.