TimeDataModel is a metadata-rich container for time series data. It lets you carry your data and its context — name, unit, frequency, timezone — as a single, self-describing object, fully interoperable with pandas, NumPy, Polars, and PyArrow.

⬇️ Installation  |  📖 Documentation  |  🚀 Examples

🧱 Core Data Classes

📐 Data Shapes

TimeSeries supports four temporal shapes to model everything from simple point-in-time data to full audit trails:

🚀 Quick Start

import pandas as pd
from timedatamodel import TimeSeries, Frequency

# --- Univariate series from a pandas DataFrame ---
df = pd.DataFrame({
    "valid_time": pd.date_range("2024-01-01", periods=24, freq="h", tz="UTC"),
    "value": [100.0 + i * 2.5 for i in range(24)],
})

ts = TimeSeries.from_pandas(
    df,
    frequency=Frequency.PT1H,
    name="wind_power",
    unit="MW",
)

print(ts)
# TimeSeries ─────────────────────────
#   Name        wind_power
#   Shape       SIMPLE
#   Rows        24
#   Frequency   PT1H
#   Timezone    UTC
#   Unit        MW
#  ──────────────────────────────────────────
#                  wind_power
#  2024-01-01 00:00   100.0
#  2024-01-01 01:00   102.5
#  ...

# --- Unit conversion (requires pint extra) ---
ts_kw = ts.convert_unit("kW")

# --- Format conversions ---
df_pd  = ts.to_pandas()       # pd.DataFrame with datetime index
df_pl  = ts.to_polars()       # pl.DataFrame
cols   = ts.to_list()         # dict[str, list] — column-oriented
arr    = ts.to_numpy()        # dict[str, np.ndarray] — column-oriented (requires numpy)
tbl    = ts.to_pyarrow()      # pa.Table (requires pyarrow)

✨ Key Features

• 🔷 Four data shapes — from SIMPLE point-in-time to AUDIT full audit history;
• 🏷️ Metadata — name, unit, frequency, timezone, data type, description on every series;
• 📋 Metadata-only mode — construct TimeSeries(df=None, …) to declare a series' structure before any data exists, for catalog/registration use;
• 🔄 Format conversions — to_pandas, to_polars, to_list, to_numpy, to_pyarrow with lazy optional-dependency checks;
• 📊 Coverage bar — coverage_bar() renders null coverage as a binned SVG in Jupyter or Unicode blocks in terminal;
• 🗺️ Geospatial primitives — GeoLocation and GeoArea for use by consumer layers;
• 📏 Units — optional pint integration for dimensional unit conversion and validation;
• ⚡ Polars-powered — backed by the Polars compute engine for high-performance in-memory processing;
• 🐍 Type-safe — full type hints with PEP 561 support.

⬇️ Installation

Install the stable release:

pip install timedatamodel

Install with optional dependencies:

pip install timedatamodel[pandas]    # pandas interop (includes pyarrow for tz-aware columns)
pip install timedatamodel[pint]      # unit conversion
pip install timedatamodel[geo]       # geospatial support (shapely)
pip install timedatamodel[all]       # all optional extras

Install in editable mode for development:

git clone https://github.com/rebase-energy/timedatamodel.git
cd TimeDataModel
pip install -e .[dev]

🤝 Contributing

Contributions are welcome! Here are some ways to contribute to TimeDataModel:

• Propose new features or extend existing classes;
• Improve documentation or add example notebooks;
• Report bugs or suggest features via GitHub Issues.

📄 Licence

This project uses the MIT Licence.