timedb 0.2.0

Minimal stateless ClickHouse library for bitemporal time series

⏱️ TimeDB

An open-source, opinionated time-series database built on PostgreSQL & ClickHouse.

TimeDB is designed to natively handle overlapping forecast revisions, auditable human-in-the-loop updates, and "time-of-knowledge" history. Using a three-dimensional temporal data model, it provides a seamless workflow through its Python SDK and FastAPI backend.

Traditional time-series databases assume one immutable value per timestamp. TimeDB is built for complex, real-world domains where forecasts evolve, data gets corrected, and historical backtesting requires strict audits.

---

🧊 The 3D Temporal Data Model

At the heart of TimeDB is its three-dimensional approach to time. We track not just when data is valid, but when it became known and when it was altered.

Dimension	Description	Real-World Example
📅 valid_time	The time the value represents a fact for.	"Wind speed forecast for Wednesday 12:00"
⏰ knowledge_time	The time when the value was predicted/known.	"Generated on Monday 18:00"
✏️ change_time	The time when the value was written or changed.	"Manually overridden on Tuesday 09:00"

Audit & Metadata: Every data point also supports tags, annotations, and changed_by to maintain a perfect audit trail of who changed what, when, and why.

---

✨ Why Choose TimeDB?

• 📊 Forecast Revisions: Store overlapping forecasts with full provenance.
• 🔄 Auditable Updates: Manual adjustments generate audit trails, not silent overwrites.
• ⏪ True Backtesting: Query historical data as of any point in time ("What did our model know last Monday?").
• 🏷️ Label-Based Organization: Easily filter and slice series by meaningful dimensions.

---

🚀 Quick Start

1. Installation

pip install timedb

Requires Python 3.9+, PostgreSQL (series metadata), and ClickHouse (time-series values).

2. Usage Example

import timedb as td
import pandas as pd
from datetime import datetime, timezone
from timedb import TimeSeries

# Create Schema
td.create()

# 1. Create a forecast series
td.create_series(
    name="wind_power",
    unit="MW",
    labels={"site": "offshore_1"},
    overlapping=True
)

# 2. Insert forecast with knowledge_time
knowledge_time = datetime(2025, 1, 1, 18, 0, tzinfo=timezone.utc)
ts = TimeSeries.from_pandas(
    pd.DataFrame({
        'valid_time': pd.date_range('2025-01-01', periods=24, freq='h', tz='UTC'),
        'value': [100 + i*2 for i in range(24)]
    }),
    unit='MW',
)

td.get_series("wind_power") \
   .where(site="offshore_1") \
   .insert(ts, knowledge_time=knowledge_time)

# 3. Read latest forecast — returns a TimeSeries
ts_latest = td.get_series("wind_power").where(site="offshore_1").read()
df_latest = ts_latest.to_pandas()  # pd.DataFrame with valid_time index

# 4. Read forecast history (one row per knowledge_time × valid_time)
ts_all = td.get_series("wind_power").where(site="offshore_1").read(overlapping=True)

---

🧪 Try it in Google Colab

Want to try TimeDB without a local setup? Open our Quickstart in Colab — the first cell automatically installs PostgreSQL + ClickHouse inside the VM.

Note: Data persists only within the active Colab session. Additional notebooks are available in the examples/ directory.

---

📚 Documentation & Resources

• 📖 Official Documentation
• ⚙️ Installation Guide
• 🐍 Python SDK Documentation
• 🌐 Reference
• 💡 Examples & Notebooks

---

🤝 Contributing

Contributions are welcome! If you're interested in improving TimeDB, please see our Development Guide for local setup instructions.

---

Licensed under the Apache-2.0 License.

Find a bug or have a feature request? Open an Issue.