Schedule
The tutorial is on Monday, August 26 starting at 14:00 in room 124-125. The speakers will be available afterwards for Q&A, as well as during the presentation via text in the sessions Whova chat.
-
Introduction to Time Series Machine Learning - (14:00, 15 Minutes, by Anthony Bagnall)
Special considerations must be given when it comes to processing and creating models from time series data. To begin we will be a brief overview of the time series classification (TSC) and time series extrinsic regression (TSER) tasks, including the datatypes used in this tutorial and the different kinds of time series data one may find.
-
Distance Measure for Time Series - (14:15, 15 Minutes, by Geoffrey Webb)
Many classifiers and regressors use distance or similarity functions between whole time series as the basis for prediction. Historically, distance functions have been mostly used with Nearest Neighbour (NN) classifiers such as 1-NN Dynamic Time Warping (1NN-DTW). We will highlight recent work in comparing time series using distances and give a hands-on demonstration of how we can define a new distance function in aeon to use with scikit-learn or aeon estimators.
-
Convolutional Kernels - (14:30, 15 Minutes, by Geoffrey Webb)
Kernel/Convolution classifiers use convolutions with kernels, which can be seen as subseries used to derive discriminatory features. Each kernel is convolved with a time series through a sliding dot product creating an activation map. We will review the ROCKET family of algorithms and use these algorithms to demonstrate parameter tuning in aeon.
-
Summary Features - (14:45, 15 Minutes, by Matthew Middlehurst)
A family of popular approaches involve extracting global features passed to a standard estimator in a simple pipeline. We will review popular feature generators such as Catch22 and TSFresh, and demonstrate how to create simple feature estimators using the aeon transformation interface and the pipeline utilities present in both aeon and scikit-learn.
-
Phase Dependent Intervals - (15:00, 15 Minutes, by Matthew Middlehurst)
Features found from the whole series may be confounded by noise. Another family of algorithms mitigate against this by finding intervals from numerous extracted subseries called intervals. In addition to demonstrating usage and providing examples of when interval methods may be useful, we will demonstrate some of the visualisation techniques available for evaluating feature importance.
-
Break (15:15, 15 Minutes)
-
Dictionary-based Bag-of-word Models - (15:30, 15 Minutes, by Ali Ismail-Fawaz)
Dictionary based approaches use histograms of counts of repeating patterns as features for the estimator. We will describe these estimators and use them to highlight different design patterns used in TSC and TSER research in terms of ensembles vs pipelines and randomisation vs optimisation.
-
Deep Learning for Time Series - (15:45, 25 Minutes, by Germain Forestier)
Since a review of deep learning classification in 2019 numerous algorithms have been proposed, such as InceptionTime and its variants. We will provide a short summary of work in this area before reviewing the basic structure of networks and how to structure a new deep learning estimator.
-
Phase Independent Shapelets - (16:10, 15 Minutes, by Anthony Bagnall)
Shapelets are phase independent discriminatory subseries taken from the training data. They have proved a popular primitive for classification. One feature of shapelets is the ability to trace important shapelets back to the training data to give domain specific insights. We will demonstrate how to simply do this in aeon.
-
Hybrid Approaches - (16:25, 15 Minutes, by Anthony Bagnall)
This type of method combines two or more of the above approaches. The nature of the data and the problem dictate which category of algorithm is most appropriate. The most accurate algorithms on average, with no a-priori knowledge of the best approach, combine multiple transformation types in a hybrid algorithm. We will demonstrate how to build a bespoke hybrid ensemble classifier using the estimators covered in the tutorial.
-
Conclusion and Future Outlook - (16:40, 10 Minutes, by Anthony Bagnall)
We conclude our tutorial, with some discussion of future work in the fields of TSC and TSER.
-
Q&A (16:50, 10 Minutes)