Dataflow notebooks are Jupyter notebooks written in Python that use a modified kernel (dfkernel) to enable new features. Specifically,

• Cells have persistent identifiers, allowing any output to be robustly referenced (e.g. Out[a32b3f])
• By leveraging the cell identifiers, any cell execution will recursively update any referenced cells that have changed just like dataflows which update upstream modules as needed
• There is no need to order cells so that dependent cells are before the cell that uses them
• Navigation is enhanced by links to referenced cells and directed graph visualizations of the induced dataflow
• Intermediate outputs are both viewable via the rich Jupyter output formats and usable in later cells
• Dataflow notebooks can be converted to standard python notebooks

Video:

Projects:

• dfkernel: a Jupyter kernel that provides dataflow features
• dfconvert: a script to transform ipynb files created using the dataflow kernel to a standard python notebook