[ENH] TS2Vec implementation

[gasperpetelin]

Describe the feature or idea you want to propose

It might be beneficial to include an implementation of TS2Vec (Towards Universal Representation of Time Series) in the library. TS2Vec is a popular and efficient contrastive learning approach for time series representation. It could be implemented as a transformer similar to Rocket or TSFresh. The original author's code is available here (MIT licence). I would be happy to add the implementation if this is something the maintainers/developers of aeon agree should be included and if it aligns with future development.

Describe your proposed solution

I believe the implementation could be similar to Rocket. At first, it could support only encoding_window='full_series', so the returned array would always have the shape n_instances x output_dim, primarily useful for classification.

Example:

trf = TS2Vec(output_dim=320)
trf.fit(X_train)
X_train = trf.transform(X_train)
X_test = trf.transform(X_test)

Describe alternatives you've considered, if relevant

No response

Additional context

Classification accuracy of TS2Vec compared to 2 other randomly selected TSC models

TS2Vec	Catch22	MultiRocket	dataset
0.84	0.85	0.9	ACSF1
0.565217	0.734015	0.828645	Adiac
0.788571	0.748571	0.874286	ArrowHead
0.986667	0.913333	1	BME
0.766667	0.6	0.766667	Beef
0.8	0.75	0.85	BeetleFly
0.8	0.9	0.9	BirdChicken
1	0.943333	0.995556	CBF
0.833333	0.766667	0.933333	Car
0.982507	0.900875	0.976676	Chinatown
0.590104	0.603906	0.7875	ChlorineConcentration
0.83913	0.792754	0.95	CinCECGTorso
1	1	1	Coffee
0.636	0.732	0.804	Computers
0.766667	0.582051	0.797436	CricketX
0.75641	0.55641	0.851282	CricketY
0.776923	0.635897	0.828205	CricketZ
0.691131	0.70625	0.78	Crop
0.977124	0.895425	0.964052	DiatomSizeReduction
0.741007	0.705036	0.76259	DistalPhalanxOutlineAgeGroup
0.735507	0.800725	0.804348	DistalPhalanxOutlineCorrect
0.683453	0.661871	0.690647	DistalPhalanxTW
0.91	0.82	0.92	ECG200
0.942667	0.937333	0.946444	ECG5000
1	0.778165	1	ECGFiveDays
0.577348	0.546961	0.640884	EOGHorizontalSignal
0.505525	0.505525	0.535912	EOGVerticalSignal
0.748201	0.748201	0.726619	Earthquakes
0.694462	0.73233	0.729737	ElectricDevices
0.384	0.364	0.626	EthanolLevel
0.833728	0.761538	0.799408	FaceAll
0.965909	0.602273	0.931818	FaceFour
0.943902	0.698049	0.96	FacesUCR
0.714286	0.593407	0.859341	FiftyWords
0.942857	0.771429	0.988571	Fish
0.928788	0.917424	0.952273	FordA
0.780247	0.750617	0.832099	FordB
0.994386	0.997895	0.999649	FreezerRegularTrain
0.960702	0.960351	0.994737	FreezerSmallTrain
0.98	0.96	1	GunPoint
0.996835	0.974684	1	GunPointAgeSpan
1	0.993671	1	GunPointMaleVersusFemale
1	1	1	GunPointOldVersusYoung
0.790476	0.609524	0.733333	Ham
0.483766	0.470779	0.512987	Haptics
0.59375	0.5625	0.71875	Herring
0.387273	0.418182	0.474545	InlineSkate
1	1	1	InsectEPGRegularTrain
1	1	1	InsectEPGSmallTrain
0.632828	0.564646	0.669192	InsectWingbeatSound
0.962099	0.878523	0.969874	ItalyPowerDemand
0.872	0.826667	0.872	LargeKitchenAppliances
0.868852	0.688525	0.721311	Lightning2
0.835616	0.671233	0.835616	Lightning7
0.941151	0.904478	0.928785	Mallat
0.933333	0.85	0.933333	Meat
0.739474	0.748684	0.801316	MedicalImages
0.62987	0.61039	0.564935	MiddlePhalanxOutlineAgeGroup
0.817869	0.769759	0.838488	MiddlePhalanxOutlineCorrect
0.564935	0.545455	0.532468	MiddlePhalanxTW
0.90433	0.920825	0.980619	MixedShapesRegularTrain
0.863093	0.857732	0.964536	MixedShapesSmallTrain
0.866613	0.879393	0.944888	MoteStrain
0.876845	0.844275	0.959796	NonInvasiveFetalECGThorax1
0.889567	0.874809	0.967939	NonInvasiveFetalECGThorax2
0.768595	0.690083	0.966942	OSULeaf
0.833333	0.766667	0.933333	OliveOil
0.778555	0.79021	0.851981	PhalangesOutlinesCorrect
0.290084	0.304325	0.357595	Phoneme
1	1	1	Plane
0.955556	0.95	0.983333	PowerCons
0.853659	0.839024	0.853659	ProximalPhalanxOutlineAgeGroup
0.883162	0.831615	0.924399	ProximalPhalanxOutlineCorrect
0.795122	0.765854	0.795122	ProximalPhalanxTW
0.578667	0.498667	0.52	RefrigerationDevices
0.421333	0.501333	0.536	ScreenType
0.93	0.933333	0.958333	SemgHandGenderCh2
0.795556	0.808889	0.773333	SemgHandMovementCh2
0.895556	0.857778	0.926667	SemgHandSubjectCh2
0.988889	0.977778	1	ShapeletSim
0.866667	0.785	0.926667	ShapesAll
0.728	0.8	0.832	SmallKitchenAppliances
0.986667	0.893333	0.986667	SmoothSubspace
0.875208	0.831947	0.886855	SonyAIBORobotSurface1
0.887723	0.919203	0.940189	SonyAIBORobotSurface2
0.963453	0.968067	0.981302	StarLightCurves
0.972973	0.92973	0.972973	Strawberry
0.9248	0.896	0.9808	SwedishLeaf
0.967839	0.965829	0.98191	Symbols
0.993333	0.983333	0.996667	SyntheticControl
0.916667	0.894737	0.951754	ToeSegmentation1
0.892308	0.776923	0.915385	ToeSegmentation2
1	1	1	Trace
0.988586	0.806848	0.998244	TwoLeadECG
0.9995	0.844	1	TwoPatterns
0.993056	0.861111	0.993056	UMD
0.907314	0.824121	0.980179	UWaveGestureLibraryAll
0.771915	0.758236	0.867672	UWaveGestureLibraryX
0.686767	0.705472	0.807091	UWaveGestureLibraryY
0.739252	0.703238	0.814908	UWaveGestureLibraryZ
0.996269	0.998053	0.999513	Wafer
0.814815	0.462963	0.851852	Wine
0.681818	0.540752	0.775862	WordSynonyms
0.662338	0.753247	0.766234	Worms
0.714286	0.844156	0.766234	WormsTwoClass
0.846	0.779333	0.92	Yoga

[TonyBagnall]

sounds like a good edition, go for it

[hadifawaz1999]

Hello, that would be a great addition of course ! it is actually on a to do list so it would be great if you are willing to do it. We already have a pull request where we're creating a sub-module of transformation calling it self_supervised. Currently implementing a model called TRILITE, for more info checkout the PR #2385

Working on the initialization of the module, however feel free to PR your code and once we have this one merged you can adjust your code to be inside the module

[gasperpetelin]

Hi. Thanks for pointing me to your PR. I pushed my changes to #2825, but it is still marked as a draft. I didn't see your PR before I started with the TS2Vec implementation, so I placed everything in contrastive_based. However, I can subsequently move it into self_supervised.