Notebook environments are used widely in data analysis and machine learning
because their interactive user interfaces fit well with exploratory tasks in
these areas. However, the execution model of existing notebook environments
is unsuitable for safe and efficient exploration because every code cell of
a notebook runs in a single execution environment even for independent
exploratory tasks, where unintended interference can arise among them. To
resolve this problem, we developed Multiverse Notebook, a notebook
environment that runs each cell in a separate execution environment and
saves its execution state. In this poster, we present Multiverse Notebook
and our techniques to reduce the application's time and space of Multiverse Notebook.