Skip to content

TimeNet

TimeNet gives time-series datasets one format and one Python API. You can build a raw source into TimeF, publish the result to a registry, and load it without dataset-specific code. The same objects represent ECGs, accelerometer traces, market series, and other sequential data.

Scope

TimeNet is not a modeling toolkit. Training, inference, model definitions, and evaluation metrics are out of scope. TimeNet only gives you the data.

How it fits together

TimeNet architecture diagram

A connector turns a raw source into a TimeF version. A registry stores immutable versions. The client searches the registry and opens datasets with lazy Signal values, or loads them with every value in memory. Consumer code never runs a connector.

  • BaseConnector is the only contract a new data source must satisfy.
  • TimeFDataset is the in-memory model a connector populates during convert().
  • TimeFWriter writes a populated dataset as one immutable version.
  • TimeFReader restores the version without loading every Signal value.

Where next

  • Get started: install TimeNet and load your first dataset.
  • Architecture: how the packages, registries, and build flow fit together.
  • Datasets: the datasets already built into the registry.