Type example — an AnnData composite type

AnnData (anndata.py) is a custom type for data that is several data objects belonging together: a single-cell experiment is an expression matrix plus per-cell annotations plus per-gene annotations, and the useful unit is the whole bundle. Where the Image example subclasses Array because its data is one array, this type subclasses CompositeData — the core type for a named bundle of slots.

What expected_slots buys you

class AnnData(CompositeData):
    expected_slots: ClassVar[dict[str, type]] = {
        "X": DataFrame,    # expression counts: one row per gene, one column per cell
        "obs": DataFrame,  # per-cell metadata: one row per cell
        "var": DataFrame,  # per-gene metadata: one row per gene
    }

A subclass that fills in expected_slots fixes the bundle's layout:

• Construction is validated. set() (and the slots= constructor) rejects a value whose type does not match its slot, so a mis-assembled bundle fails at the block that built it, not three steps downstream.
• Ports can demand the whole. A block that declares AnnData on an input port receives all three tables together and can rely on each slot's type.

bundle = AnnData(slots={"X": counts, "obs": cells, "var": genes})
cells = bundle.get("obs").to_pandas()      # slots read like named attributes

Composites elsewhere in the product

Composite types are also how you plot two outputs through one port: a plot binds to exactly one output, so a figure that needs an embedding from one step and cluster labels from another starts by combining them into a composite type and a merge block (see the writing-plots.md page).

Completing the type

As with every type, pair it with a SimpleLoader so the Load block can read it — the IO block example shows the pattern.