Practical overview
When it fits
Use tsfresh when offline sequential data is grouped by sample and you want to calculate a broad range of features systematically before analysis or modeling.
Who it is for
- Python practitioners preparing sequential or time-series data for analysis.
- Teams building classification or regression workflows that need automated feature generation and selection.
- Analysts using pandas or scikit-learn alongside time-series feature extraction.
Documented uses
- Generate numerical characteristics from grouped sequential measurements.
- Select features relevant to a target label or prediction task.
- Prepare feature data for downstream classification or regression with a separate machine-learning package.
- Describe time-series dynamics or prepare features for clustering.
Requirements & limits
- tsfresh extracts features but does not train the downstream machine-learning model; the documentation points to scikit-learn for model training.
- The documentation says it is not suitable for streaming operations.
- For highly irregular series, some features assume equal time spacing even though timestamps are used to order observations.
- The quick start documents a Dask installation extra for large time series that may not fit in memory.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install tsfresh and follow the documented robot-failure example
Set up the Python package and follow the quick-start stages for extracting and selecting features from grouped sensor time series.
Before you start
- A Python environment with pip.
- The example uses pandas data with an observation ID column, a time-order column, and a target label for each ID.
Steps
Install the package with the command below.
pip install tsfreshLoad the robot execution failures example dataset and its target labels using the example loader documented in the quick start.
Extract features with
extract_features, using the ID column to group series and the time column to order observations.Impute missing values and select target-relevant features with
imputeandselect_features.
Expected outcome
A feature DataFrame selected against the example’s failure labels, ready to be used with a separate model-training workflow.
Guide limitations
- The documented example treats model training as a separate step, using scikit-learn.
- The quick start notes that some feature calculators produce NaN values and demonstrates imputing them before feature selection.
- tsfresh is not intended for streaming data; features that assume equal time spacing need care with highly irregular series.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 12 October 2026
- Original website title
- tsfresh — tsfresh 0.21.2.post0.dev2+g6af3c00 documentation · Official homepage title · retained in its source language
Related options
hctsa — The introduction identifies this MATLAB package as another option for automatic time-series feature extraction and notes that it can be used from Python through pyopy.
featuretools — The introduction lists featuretools among other available packages.
Explore project →FATS — The introduction lists FATS among other available packages.
cesium — The introduction lists cesium among other available packages.
Explore project →