Classifier
MrSQM
MrSQM = Multiple Representations Sequence Miner.
Direct Interface to MrSQMClassifier from mrsqm. Note: mrsqm itself is copyleft (GPL3). This interface is permissive license (BSD3).
MrSQM is an efficient time series classifier utilizing symbolic representations of time series. MrSQM implements four different feature selection strategies = (R,S,RS,SR) that can quickly select subsequences from multiple symbolic representations of time series data.
Quickstart
python
from sktime.classification.shapelet_based import MrSQM
estimator = MrSQM(strat='RS', features_per_rep=500, selection_per_rep=2000, nsax=1, nsfa=0, custom_config=None, random_state=None, sfa_norm=True)Parameters(8)
- stratstr, one of ‘R’,’S’,’SR’, or ‘RS’, default=”RS”
- feature selection strategy. By default set to ‘RS’. R and S are single-stage filters while RS and SR are two-stage filters.
- features_per_repint, default=500
- (maximum) number of features selected per representation.
- selection_per_repint, default=2000
- (maximum) number of candidate features selected per representation. Only applied in two stages strategies (RS and SR), otherwise ignored.
- nsaxint, default=1
- number of representations produced by sax transformation.
- nsfaint, default=0
- number of representations produced by sfa transformation.
- custom_configdict, default=None
- customized parameters for the symbolic transformation.
- random_stateint, default=None.
- random seed for the classifier.
- sfa_normbool, default=True.
- whether to apply time series normalisation (standardisation).
References
[1]
Thach Le Nguyen and Georgiana Ifrim. “MrSQM: Fast Time Series Classification with Symbolic Representations and Efficient Sequence Mining” arXiv preprint arXiv:2109.01036 (2021).
[2]
Thach Le Nguyen and Georgiana Ifrim. “Fast Time Series Classification with Random Symbolic Subsequences”. AALTD 2022.