I don't like notebooks.- Joel Grus (Allen Institute for Artificial Intelligence)

Caption notes

In this 2018 JupyterCon talk, Joel Grus appreciates how notebooks put prose, code, and plots together. He also identifies an easily missed problem: the order of cells does not necessarily reflect their execution history, and edited code may not have been run again. He distinguishes presenting a result from reproducing an analysis. To help others reproduce and extend an analysis, he advocates recording dependencies and parameters explicitly and moving reusable logic into independently testable modules. In the Q&A, he also sees value in exploration that appends cells in order.