04 reload jupyter kernels periodically

Jupyter keeps everything in the global scope.

Every variable or function defined during a session stays in memory until explicitly deleted or the kernel is reloaded. Each imported module and function also remains in the scope.

Because of that, it’s a good habit to reload the kernel regularly. Especially, before committing results to version control or running a scheduled job. Otherwise, there are plenty of opportunities to get into trouble:

• A function with a local variable whose name matches a variable defined later in the notebook gets overridden after the function’s cell is re-evaluated.

• A more general case: earlier cells are affected by state introduced in later cells.

• Functions and variables from deleted cells still roam the scope, becoming invisible phantoms — especially short-named ones like x or cnt, leading to very annoying bugs.

• Outputs produced by older versions of code, leading to unexpected results.

Of course, this list isn’t complete.