A lot of the calculations that I do are pretty heavy and requires loading a lot of data. Currently, ObservableHQ evicts this when reloading. It would be great if it was persisted between reloads (localstorage or something?) as this is one of the nicest thins with Python notebooks.
Cells often produce values that are not serializable—DOM elements with bound event listeners, closures that capture local variables, etc.—so without a new browser API to take snapshots of the JavaScript runtime, it won’t be possible to restore the state of a notebook automatically. (See the structured clone algorithm for more on the types of things that can be serialized.)
But, we would like to offer some sort of persistent storage / data integration in the future, so that you have a place to save and restore transformed data rather than needing to re-transform it every time the notebook loads. In the interim, probably your best bet is to download your transformed data using DOM.download and then hosting it on your GitHub account as described here:
https://beta.observablehq.com/@mbostock/introduction-to-data