Get started¶

This guide builds a small offline dataset and loads it through the Python client.

Install the client¶

TimeNet needs Python 3.11 or newer. Install the core package, then add only the extras that you use.

uv add timenet
uv add 'timenet[cli]'    # add the timenet command
uv add 'timenet[torch]'  # add the PyTorch adapter

pip install timenet
pip install 'timenet[cli]'
pip install 'timenet[torch]'

The torch extra accepts any PyTorch build. Install a CPU or accelerator build that matches your environment before you add the extra.

Install the build tool¶

The timenet-build command ships in the separate connectors package. Install it as an isolated tool:

uv tool install timenet-connectors

Use pipx install timenet-connectors if you use pipx for command-line tools.

Build the example dataset¶

Build the offline timenet/hello-world dataset into a local registry:

timenet-build build timenet/hello-world --out .timenet-registry

The command needs no network access. It writes this version:

.timenet-registry/timenet/hello-world/1.0.0/
├── manifest.json
├── control.duckdb
└── time_series/

Load the dataset¶

Pass the same registry path to the client:

import pandas as pd

from timenet.client import TimeNet

client = TimeNet(".timenet-registry")
dataset = client.load("timenet/hello-world")
dataset.describe()

signal = dataset.records[0].signals[0]
frame = pd.DataFrame({signal.name: signal.to_numpy()})
print(frame.head())

load() reads the whole dataset with every value in memory. For a dataset too large for that, client.open() returns a reader whose reads leave Signal values in storage until to_arrow(), to_numpy(), or read_steps() reads them; see Client.

Continue with these pages:

• Client covers search, version pins, local builds, and PyTorch.
• Data model explains Records, Sources, Signals, Annotations, and Tasks.
• Connectors explains how to add a dataset.
• Build and publish explains local and remote registries.