Skip to main content

Crate somatize_data

Crate somatize_data 

Source
Expand description

Where the data comes from, and what it is once it arrives.

A source is a node, and there is no Source trait: a source takes something and answers with something, which is Node, and a second trait with a method that does what forward does is both a hole with one tenant and the E0034 the rules warn about. Being a node is the point — .at(), .cached(), .mapped(), the record and the figure all reach a dataset with nothing written here.

What the graph is handed is a coordinate and not a batch, because the input is the one value a cache hashes by content. Measured on 24 August 2026, release build, one node returning a constant so only the input grows:

what is handed to forwardwith a store behind it
1 MB of tensor6,1 ms
19 MB of tensor (32×3×224×224)121 ms, every step, hit or miss
a Span0,027 ms

Nothing there is pathological — torch.save is 1 ms/MB and sha256 2 ms/MB — which is why the answer is not a faster hash.

The other half of the name is the dataset’s version, and a source has to state it without reading itself. Against a Store that is free: a name resolves to a digest and the digest is the content hash, so Parquet::version costs one resolve and no bytes.

Not here: SQL, decided rather than delayed, since every Rust driver worth using carries a runtime and Store is synchronous on purpose; and ranged reads, which need the store to learn to read a range first.

Structs§

Frame
Rows and columns, together with what each column is called and holds.
FrameError
Why that could not be written down, or read back.
Ipc
Writes frames down so they can be kept or sent, and reads them back.
Parquet
A parquet file kept in a store, answering spans of rows.
ParquetError
Why those rows could not be had.
Span
The rows a source is being asked for: take of them, starting at at.
SpanError
Why that was not a span.