blue-yonder/tsfresh
Tsfresh
tsfresh is an automated feature engineering tool and library designed to extract statistical characteristics from raw time series data. It transforms sequential data into tabular datasets, converting time series into a flat format where each row represents a unique entity and columns represent extracted features.
The project distinguishes itself through a parallel data processing framework that distributes heavy computational workloads across multiple CPU cores. It also implements hypothesis-based feature selection to identify the most predictive characteristics and filter out irrelevant ones to reduce dimensionality.
The system covers a broad range of machine learning preprocessing capabilities, including missing value imputation, rolling window frame generation for forecasting, and the integration of extraction and selection steps into a single reproducible pipeline.
Features
- Statistical Feature Extractors - Calculates comprehensive statistical characteristics from raw time series sequences to create tabular datasets for machine learning.
- Automated Feature Selection Tools - Automatically identifies and retains only the most predictive characteristics to reduce dimensionality and prevent overfitting.
- Forecasting Input Preparation - Formats temporal data into rolling window sets for training and evaluating forecasting models.
- Irrelevant Feature Filters - Removes characteristics that do not contribute to the target task by evaluating explaining power via hypothesis testing.
- Tabular Converters - Transforms raw time series sequences into flat tabular dataframes suitable for standard machine learning models.
- Hypothesis Testing Frameworks - Implements rigorous statistical tests, such as p-value calculations, to validate the significance of extracted features.
- Time Series Forecasting - Prepares temporal data sequences for prediction models by transforming them into rolling window frames.
- Time Series Machine Learning - Preprocesses temporal and sequential data through imputation and rolling window generation for machine learning.
- Feature Engineering Tools - Provides an automated system for transforming raw sequence data into predictive features through statistical calculations.
- Parallel Processing - Distributes heavy statistical feature extraction workloads across multiple CPU cores to process large time series datasets efficiently.
- Time-Series Statistical Profiling - Extracts complex statistical characteristics from time-indexed data using optimized vectorized operations.
- Time Series Toolkits - Offers a comprehensive toolkit for automatically extracting statistical characteristics from raw time series data.
- Forecasting Window Generators - Generates overlapping rolling window frames from sequential data to prepare training sets for time-dependent forecasting.
- Large-Scale Data Computation - Processes massive time series datasets by distributing heavy computational workloads across multiple CPU cores.
- Custom Parallel Task Execution - Distributes heavy feature calculation workloads across multiple CPU cores using a worker-based task queue.
- Machine Learning Pipelines - Integrates data extraction and filtering steps into a unified workflow for simplified model training and deployment.
- Compositional Transformation Pipelines - Chains feature extraction, imputation, and selection into a single reproducible sequence for preprocessing.
- Feature Engineering - Automatic extraction of features from time series data.
- Time Series Analysis - Automatic feature extraction for time series datasets.
- Feature Engineering - Automatic feature extraction from time series data.
- Time Series Analysis - Automated feature engineering for time series data.
- Automated Machine Learning - Automatic extraction of relevant features from time series data.
