Resources
Access our resources
Developed by leading experts, our repository of software and data is available to support your research and initiatives.
Benchmark data archives
-
Epilepsy Ecosystem
A crowd-sourcing ecosystem for publicly sharing data and improving the performance of seizure prediction algorithms – making seizure prediction viable for people with epilepsy.
-
TSER: Time Series Extrinsic Regression Repository
A website that supports TSER research, a regression task that aims to learn the relationship between a time series and a continuous scalar variable which closely relates to time series classification that focuses on establishing the relationship between a time series and a categorical class label.
-
Monash Forecasting Repository
A repository containing datasets of related time series for global forecasting of areas such as Bitcoin, vehicle trips, weather, electricity, car parts and renewables.
-
MONSTER: Monash Scalable Time Series Evaluation Repository
A collection of large datasets for time series classification.
Software
-
Tempo
A machine learning library for time series analytics focusing on efficient implementations of elastic distance measures.
-
Rocket
A fast and highly-accurate method of classifying time series databy transforming the input time series using a large number of random filters, and then using the outputs to train a simple linear classifier to make predictions.
-
MiniRocket
An evolution of ROCKET that combines a small set of fixed (non-random) convolutional kernels and optimisations to drastically reduce overall computational cost, making it up to 75x faster than ROCKET while maintaining similar accuracy.
-
Hydra
A combination of the key ideas from ROCKET with so-called dictionary methods, organising kernels into groups (dictionaries) and counting the most prominent patterns (represented by the kernels) at each timestep within each group. This allows for fast and accurate time series classification, while removing the need for bias values (per ROCKET and MiniRocket).
-
QUANT
A very simple approach to time series classification, recursively dividing the input time series in half, and using the quantiles of the resulting subseries (intervals) as features.
-
TS-CHIEF
A machine-learning method for classifying time series data that uses an ensemble of decision trees and distance measures to detect patterns. It combines different ways of representing time-series patterns, achieving high accuracy while remaining scalable for large datasets.
-
Proximity Forest
A machine-learning method for classifying time series data that builds many decision trees and splits the data based on similarity. This allows it to detect patterns accurately while still scaling well to very large datasets.
-
TSER: Time Series Extrinsic Regression
A framework and codebase for predicting a numeric value from a time series dataset using machine learning models. It provides datasets, feature transformations and models so researchers can test and compare different approaches to time series regression.
-
FastEE: Fast Ensembles of Elastic Distances
A machine-learning method for time-series classification that speeds up the training of an elastic-distance ensemble classifier – keeping similar accuracy to the original method but training much faster, making it more practical for larger datasets.
-
FastWWSearch: Fast warping window search algorithm for DTW
A machine‑learning tool for time series classification that quickly finds the best warping window for dynamic time warping.
-
Tight DTW Lower Bounds
A collection of fast lower‑bound methods that speed up Dynamic Time Warping calculations for time‑series similarity.
-
LbEnhanced
A collection of improved lower‑bound techniques for Dynamic Time Warping (DTW) that help time‑series algorithms speed up similarity calculations. These enhanced bounds make it quicker to rule out unlikely matches so fewer full DTW comparisons are needed.
-
TSI: Time Series Indexing
A tool for quickly searching and retrieving similar patterns in large time‑series datasets. It builds a structure that lets you compare and find matching time series within a large set efficiently.
-
Skopus: Mining top-k sequential patterns under leverage
A time‑series pattern mining tool that finds the most interesting sequential patterns (sub‑sequences that occur often) in sequential data using a fast exact search algorithm. It’s designed to identify top‑k patterns that are more meaningful than would be expected by chance, so you can discover important recurring sequences efficiently.
-
DBA: Dynamic Time Warping Barycenter Averaging
A method for computing an average pattern from a set of time‑series sequences by aligning them using Dynamic Time Warping. This produces a representative ‘average’ time‑series that respects the timing differences between samples, which is useful for clustering and summarising time‑series data.
-
Temporal Convolutional Neural Network
A deep learning method that uses convolutional neural networks to automatically learn and classify patterns in time‑series data.
-
NeuralProphet
A time‑series forecasting framework that blends traditional trend/seasonality models with neural networks for flexible, interpretable predictions.
-
Recurrent Neural Network Implementations for Time Series Forecasting
A set of example code showing how to use Recurrent Neural Networks (RNNs) like LSTMs to forecast future values from time‑series data (i.e. to learn patterns over time and predict what comes next).
-
Rlgt
An R package for time‑series forecasting that implements Bayesian extensions of exponential smoothing models with trend and seasonality adjustments, letting you make flexible forecasts that can handle different error behaviours and trends.
-
Calibration by Constrained Transformation
A monotonic post-hoc calibration method that corrects miscalibrated classifier probabilities without altering predictions, using a constrained transformation on sorted logits that is interpretable, data-efficient, and achieves state-of-the-art calibration performance across a range of models and datasets.