Cells can evaluate out of order

One common pitfall of the notebook is that it is stateful. Once a cell is executed, its value remains until it is re-evaluated. This can lead to unreproducible notebooks if due care is not taken.

Notebooks frequently contain multiple cells, often describing the sequential steps taken to process some data. However, as the data is being explored, the data scientist may edit earlier cells to fix an error or modify the data format.

doing this produces an inconsitent state, as the notebook is not re-evaluated when a single cell is modified. This can result in the notebook being unreproducible, as some cells may have been computed on data that no-longer exists.

The workaround

Whenever the data scientist modifies a code cell, they should manually re-evalute the entire notebook. This ensures that the cells always have valid and reproducible data in them.