timedatamodel 0.1.6

A lightweight data model for time series data with pandas, numpy, and polars support

TimeDataModel

A lightweight Python data model for time series data, built on Polars.

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

⬇️ Installation  |  📖 Documentation  |  🚀 Examples

---

🧱 Core Data Classes

Class	Description
📈 TimeSeriesPolars	Univariate time series backed by a Polars DataFrame, supporting four temporal shapes
📊 TimeSeriesTablePolars	Multivariate time series — multiple named columns sharing the same valid_time index
🔷 DataShape	Enum that selects which timestamp columns are present: SIMPLE, VERSIONED, CORRECTED, or AUDIT
⏱️ Frequency	ISO 8601 duration-based frequencies (PT1H, P1D, P1M, …)
🏷️ DataType	Hierarchical taxonomy: ACTUAL → OBSERVATION, DERIVED; CALCULATED → FORECAST, SIMULATION, …
🗺️ GeoLocation / GeoArea	Geographic point and polygon types with distance, bearing, and containment

---

📐 Data Shapes

TimeSeriesPolars supports four temporal shapes to model everything from simple point-in-time data to fully bi-temporal audit trails:

Shape	Columns	Use case
SIMPLE	valid_time, value	Standard time series
VERSIONED	knowledge_time, valid_time, value	Bi-temporal: track when each value was produced
CORRECTED	valid_time, change_time, value	Corrections: track when a value was revised
AUDIT	knowledge_time, change_time, valid_time, value	Full audit trail

---

🚀 Quick Start

import pandas as pd
from timedatamodel import TimeSeriesPolars, TimeSeriesTablePolars, DataShape, 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 = TimeSeriesPolars.from_pandas(
    df,
    shape=DataShape.SIMPLE,
    frequency=Frequency.PT1H,
    name="wind_power",
    unit="MW",
)

print(ts)
# TimeSeriesPolars ─────────────────────────
#   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()        # structured np.ndarray (requires numpy)
tbl    = ts.to_pyarrow()      # pa.Table (requires pyarrow)

# --- Multivariate table — see examples/nb_02_timeseries_table_polars.ipynb ---

---

✨ Key Features

• 🔷 Four data shapes — from SIMPLE point-in-time to AUDIT full bi-temporal history;
• 🏷️ Rich metadata — name, unit, frequency, timezone, data type, location, labels, description on every series;
• 📊 Multivariate tables — TimeSeriesTablePolars groups co-indexed series with per-column metadata;
• 🔄 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 — attach locations, filter by radius or area, find nearest columns;
• 📏 Units — optional pint integration for dimensional unit conversion;
• ⚡ Polars native — all internal operations use the Polars compute engine;
• 🐍 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]

---

📓 Examples

#	Notebook	Topic
01	TimeSeriesPolars	Creating, inspecting, and operating on univariate polars-backed time series
02	TimeSeriesTablePolars	Multivariate tables with per-column metadata and spatial filtering

---

🤝 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.