A unified interface for machine learning with time series

🚀 Version 0.27.0 out now! Check out the release notes here.

sktime is a library for time series analysis in Python. It provides a unified interface for multiple time series learning tasks. Currently, this includes time series classification, regression, clustering, annotation, and forecasting. It comes with time series algorithms and scikit-learn compatible tools to build, tune and validate time series models.

Overview
Open Source
Tutorials
Community
CI/CD
Code
Downloads
Citation

Questions and feedback are extremely welcome! We strongly believe in the value of sharing help publicly, as it allows a wider audience to benefit from it.

Our objective is to enhance the interoperability and usability of the time series analysis ecosystem in its entirety. sktime provides a unified interface for distinct but related time series learning tasks. It features dedicated time series algorithms and tools for composite model building such as pipelining, ensembling, tuning, and reduction, empowering users to apply an algorithm designed for one task to another.

sktime also provides interfaces to related libraries, for example scikit-learn, statsmodels, tsfresh, PyOD, and fbprophet, among others.

For troubleshooting and detailed installation instructions, see the documentation.

• Operating system: macOS X · Linux · Windows 8.1 or higher
• Python version: Python 3.8, 3.9, 3.10, 3.11, and 3.12 (only 64-bit)
• Package managers: pip · conda (via conda-forge)

Using pip, sktime releases are available as source packages and binary wheels. Available wheels are listed here.

pip install sktime

or, with maximum dependencies,

pip install sktime[all_extras]

For curated sets of soft dependencies for specific learning tasks:

pip install sktime[forecasting]  # for selected forecasting dependencies
pip install sktime[forecasting,transformations]  # forecasters and transformers

or similar. Valid sets are:

• forecasting
• transformations
• classification
• regression
• clustering
• param_est
• networks
• annotation
• alignment

Cave: in general, not all soft dependencies for a learning task are installed, only a curated selection.

You can also install sktime from conda via the conda-forge channel. The feedstock including the build recipe and configuration is maintained in this conda-forge repository.

conda install -c conda-forge sktime

or, with maximum dependencies,

conda install -c conda-forge sktime-all-extras

(as conda does not support dependency sets, flexible choice of soft dependencies is unavailable via conda)

from sktime.datasets import load_airline
from sktime.forecasting.base import ForecastingHorizon
from sktime.forecasting.theta import ThetaForecaster
from sktime.split import temporal_train_test_split
from sktime.performance_metrics.forecasting import mean_absolute_percentage_error

y = load_airline()
y_train, y_test = temporal_train_test_split(y)
fh = ForecastingHorizon(y_test.index, is_relative=False)
forecaster = ThetaForecaster(sp=12)  # monthly seasonal periodicity
forecaster.fit(y_train)
y_pred = forecaster.predict(fh)
mean_absolute_percentage_error(y_test, y_pred)
>>> 0.08661467738190656

from sktime.classification.interval_based import TimeSeriesForestClassifier
from sktime.datasets import load_arrow_head
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

X, y = load_arrow_head()
X_train, X_test, y_train, y_test = train_test_split(X, y)
classifier = TimeSeriesForestClassifier()
classifier.fit(X_train, y_train)
y_pred = classifier.predict(X_test)
accuracy_score(y_test, y_pred)
>>> 0.8679245283018868

There are many ways to join the sktime community. We follow the all-contributors specification: all kinds of contributions are welcome - not just code.

• By the community, for the community -- developed by a friendly and collaborative community.
• The right tool for the right task -- helping users to diagnose their learning problem and suitable scientific model types.
• Embedded in state-of-art ecosystems and provider of interoperable interfaces -- interoperable with scikit-learn, statsmodels, tsfresh, and other community favorites.
• Rich model composition and reduction functionality -- build tuning and feature extraction pipelines, solve forecasting tasks with scikit-learn regressors.
• Clean, descriptive specification syntax -- based on modern object-oriented design principles for data science.
• Fair model assessment and benchmarking -- build your models, inspect your models, check your models, and avoid pitfalls.
• Easily extensible -- easy extension templates to add your own algorithms compatible with sktime's API.