darts

What it is and what it does

Darts is a time series forecasting and anomaly detection library that unifies classical statistical models (ARIMA, exponential smoothing) and modern deep learning approaches under a single scikit-learn-compatible interface. It handles univariate and multivariate series, supports probabilistic forecasting with confidence intervals, and enables training on multiple series simultaneously.

The library's anomaly detection module wraps forecasting models and integrates PyOD scorers to detect outliers in time series. It also supports external covariates (past-observed and future-known), static metadata, hierarchical reconciliation, and conformal prediction for calibrated uncertainty quantification. The core TimeSeries object abstracts data handling, making it straightforward to preprocess, split, and backtest models.

Use it for:

• Forecast demand, sales, or resource usage over time using ARIMA, exponential smoothing, or neural networks.
• Detect anomalies in sensor data, system metrics, or financial time series using k-means or other scorers.
• Train a single model on multiple related time series (e.g., sales across stores) to improve generalization.
• Generate probabilistic forecasts with quantile intervals for risk-aware decision-making.
• Combine predictions from multiple models to reduce forecast error through ensemble methods.
• Incorporate external variables (weather, promotions) as past or future covariates to improve predictions.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Darts provides forecasting and anomaly detection for time series using a unified scikit-learn-style API, supporting classical models like ARIMA, deep neural networks, multivariate series, and probabilistic predictions.

Yes. Darts is actively maintained, widely used (top 15000 on PyPI), has no known vulnerabilities, and offers a comprehensive, production-ready toolkit for time series forecasting and anomaly detection. The scikit-learn-style API and support for both classical and deep learning models make it suitable for both prototyping and deployment. The 18 dependencies are standard ML libraries, not exotic or risky.

Install

darts on PyPI

pip install darts

uv add darts

poetry add darts

Installing darts

Before you install

Low friction installation with a pure-Python wheel. Active maintenance with recent releases; 18 runtime dependencies including numpy, pandas, scikit-learn, and scipy suggest a mature ML stack. Requires Python 3.10+.

License in practice

Apache-2.0 (permissive) allows commercial and private use with minimal restrictions; you must include a copy of the license and state significant changes.

Quickstart

pip install darts

from darts import TimeSeries
from darts.models import ExponentialSmoothing
import pandas as pd

df = pd.read_csv("data.csv")
series = TimeSeries.from_dataframe(df, "time_col", "value_col")
train, val = series[:-36], series[-36:]

model = ExponentialSmoothing()
model.fit(train)
prediction = model.predict(len(val))

Requires Python 3.10 or later; numba and llvmlite are runtime dependencies that may require a C compiler on some systems.

Verify before relying

• Whether all 18 runtime dependencies are strictly required for basic forecasting or if some are optional for specific models.
• Performance characteristics when training on large multivariate datasets or with deep learning models.

Package facts

Evidence: darts-0.46.1-py3-none-any.whl

Keywords: time series, forecasting

Tags

More Software Development packages