Building a figure across processes

A Figure isn’t itself something a joblib/multiprocessing worker can share with the process that owns it – pickling one to hand it to a worker (or back) always produces a copy, never a live reference, so mutating an axes inside a worker never touches the original. fig.adopt_axes() is the fix: it merges an axes built standalone – most often a worker’s returned copy – into the real figure, in place of whichever of its own axes shares that grid position.

Fitting and plotting across processes, from a lazy parquet scan