Skip to content

Concepts

This page collects the vocabulary that appears across these docs. Follow a link for the full detail.

Term What it is
TimeF The one format for every dataset. It has one on-disk shape and one API, so an ECG, an accelerometer trace, and a market series all read the same way.
Dataset A named, versioned collection of records in TimeF, addressed as org/name (for example chengsenwang/tsqa).
Record One recording in a dataset (for example, a single patient trace): its time series, tasks, and annotations.
Time series One logical stream within a record, with shape (n_steps, *value_shape) and a dtype declared by its spec. You read values on demand with to_arrow(), to_numpy(), or read_steps().
Manifest The compiled manifest.json for a dataset version: the card's metadata plus the schema derived from the data. The single source of truth the SDK reads.
Connector One BaseConnector per dataset. download() fetches the raw source. convert() builds a TimeFDataset. It knows nothing about the engine or registry.
Engine run_pipeline: drives any connector through the fixed download -> convert -> derive_schema -> store pipeline, and owns caching and idempotency.
Build Running a connector through the engine to compile a dataset and publish it to a registry, via the timenet-build CLI.
Registry A served location of compiled TimeF versions that the SDK reads. It serves a Parquet control plane plus a Parquet or Zarr values plane. It never runs connector code. It can be a local directory, S3, or a remote host.
Client (SDK) TimeNet: the consumer entry point to search, download, and load datasets.
Version An immutable snapshot of a dataset, addressed org/name@version. Once its manifest.json lands, TimeNet commits it atomically.