dagpy 0.1

Data science collaboration tool based on iPython notebooks.

DAGpy is a data science collaboration tool based on iPython notebooks enabling data science teams to:

• easily collaborate by branching out of others’ notebooks

• minimize code duplication

• give a clean overview of the project

• cache intermediate outputs so team members can use them without re-evaluation

• automate the process of code execution upon data changes or on schedule

• provide a clean interface to the data visualization dashboard designers and developers

DAGpy manages a DAG (directed acyclic graph) of blocks of code, with each block being a sequence of iPython notebook cells, together with their outputs. It is designed to work seamlessly with popular VC systems like git and can be run locally or as a server application.

Author: Ivan Bestvina

Example project

To play around with the example project, you can:

• view the project DAG: python program.py view

• run all the blocks: python program.py execute -a

• add blocks through flows (with block B as a parent) and run them automatically: python program.py makeflow B -r

• commit the changes: python program.py submitflow dagpy_flow.ipynb

• explore other DAGpy options with python program.py -h

Please note that notebook execution time includes a significant overhead of over a second, because a kernel must be started for each one. In future, we plan on adding support for non-notebook plane python blocks. These would also be edited through a flow notebook view, but would be saved as .py scripts, and executed without noticable overhead.

Dependencies:

• python 3

• jupyter

• dill

• networkx, matplotlib (for DAG view)