MiniRocket

Abstract

Rocket achieves state-of-the-art accuracy for time series classification with a fraction of the computational expense of most existing methods by transforming input time series using random convolutional kernels, and using the transformed features to train a linear classifier. We reformulate Rocket into a new method, MiniRocket. MiniRocket is up to 75 times faster than Rocket on larger datasets, and almost deterministic (and optionally, fully deterministic), while maintaining essentially the same accuracy. Using this method, it is possible to train and test a classifier on all of 109 datasets from the UCR archive to state-of-the-art accuracy in under 10 minutes. MiniRocket is significantly faster than any other method of comparable accuracy (including Rocket), and significantly more accurate than any other method of remotely similar computational expense. CCS CONCEPTS• Computing methodologies → Machine learning; • Information systems → Data mining.

Predicting the outcome for COVID-19 patients by applying time series classification to electronic health records

“…The preprocessed data was used as input for MiniRocket [ 13 ] algorithm, a method for time series classification. This method uses random convolutional kernels to transform time series and uses the transformed time series as input for a linear classifier that does the actual prediction.…”

“…There are several state-of-the-art methods for time series classification, such as LSTM-FCN, cBOSS, Proximity Forest, Canonical Interval Forest (CIF), Temporal Dictionary Ensemble (TDE), InceptionTime, Rocket , TS-CHIEF, HIVE-COTE/TDE, etc. MiniRocket authors conducted benchmarks [ 13 , 14 ] with the popular UCR Time Series Classification Archive [ 15 ] datasets, including long time series and datasets with a high number of instances. The authors observed better accuracy than most of the state-of-the-art methods above mentioned.…”

“…However, MiniRocket used only a little fraction of the execution time when compared to other methods. For example, the computing time spent with the training and testing with 109 UCR datasets is more than two weeks for TS-CHIEF and eight minutes for MiniRocket [ 13 ]. Additionally, MiniRocket is compatible with both univariate and multivariate time series, there are no parameters to adjust, data normalization is not required and the results are almost deterministic.…”

“…These features are called PPV (proportion of positive values). Using the PPVs obtained for each time series, a ridge regression classifier is used to predict a class [ 13 ]. In this study we consider two classes: a positive class representing a patient’s discharge and a negative class representing a patient’s death.…”

Automatic Sleep Stage Classification with Optimized Selection of EEG Channels

“…The average age of the seven subjects is 22.4 ± 0.8 years, with the age range of 22-24, and the number of epochs totals to 6955. It includes 3 males, with average age of 22.0 ± 0.0 years and age range (22), and 4 females, with average age of 22.8 ± 1.0 years, and age range of 22-24. The EEG channels are located according to the biosemi128 configuration 1 .…”