[DOC] Shapelet transform notebook (#2046)

* Initial notebook structure

* SAST only works on univariate - will need different dataset

* made notebook framework up to viz

* collected information for gunpoint dataset

* no longer want to explore the default transformer params, rather use the same params across them where possible

* generated viz for gunpoint (not looking as expected atm)

* grouped the graph viz more clearly

* visualising all shapelets on one graph

* cleaned up and reordered

* Time Series Data On Graph

* Extended daniele's graph to be one for each class

* Added notebook plan

* showing pandas head for each transform - need descriptions

* few correction, pushing for daniele

* changed wrong docstring param naming from classifier to estimator for consistency

* Made RF classifier for each transform to rank shapelets

* Addition of RidgeClassifierCV

* Made RF classifier for each transform to rank shapelets

* Changes from random state to seed for SAST

* found a good random seed for shapelet gen

* making new branch for a quick PR

* plotting shapelets by class

* showing best and worst shapelet for a class using rccv

* trying to set up dev version

* was thinking about worst shapelets incorrectly this whole time

* gen best and worst shapelet for each class - RST

* shapelet attribute is completely different in rdst

* can't access shapelet start point, making seperate PR for this

* notebook command for reloading aeon

* local changes can now be seen in nb

* added shapelet startpos (wont commit those changes tho)

* added startpos to shapelets list

* added docstring for startpoints

* changed startpos vals to int32

* plotting dilated shapelets!

* not using both classifers for remaining transforms

* plotting dilated shapelets split into classes

* continued work on rdst

* added dilation legend to RDST shapelet plot

* _get_shp_importance handles RDST weirdly

* visualise best shapelet one class works wrong for rdst - making issue and PR

* dont need best and worst for each class as theyre mirrors

* giving RDST same length param as RST

* merge

* trying to plot the shapelets extracted with SAST

* adding analysis & trying to understand SAST

* plotted top 10 shapelets sast without start pos

* plotting SAST shapelets and understanding their roots

* using original dast attribute for source ts

* using original dast attribute for source ts

* generated all figures for notebook

* formalising notebook & adding analysis

* added notes from SAST paper

* added notes from rdst

* almost done ST analysis, trying to make sense of importance

* changed rccv to logistic regression which is easier to interpret

* finished with RST analysis

* my label encoding understanding for STC is wrong

* STC and RDST graphs and classes make sense - issue2016

* writing analysis of STC and RDST graphs

* first draft of notebook done

* article 1 for medium

* article 3 for medium

* article 2 for medium

* medium article 4

* summary of transforms

* Notebok draft ready for review

* updated aeon.registry to utils.discovery

* resolving conflicts

---------

Co-authored-by: Moonzyyy <[email protected]>

Author: IRKnyazev

aeon/classification/compose/_pipeline.py

@@ -42,7 +42,7 @@ class ClassifierPipeline(BaseCollectionPipeline, BaseClassifier):
         the str is used to name the transformer.
         The objecst are cloned prior, as such the state of the input will not be
         modified by fitting the pipeline.
-    classifier : aeon or sklearn classifier
+    estimator : aeon or sklearn classifier
         A classifier to use at the end of the pipeline.
         The object is cloned prior, as such the state of the input will not be modified
         by fitting the pipeline.

aeon/classification/shapelet_based/_rsast.py

@@ -32,7 +32,7 @@ class RSASTClassifier(BaseClassifier):
         the number of reference time series to select per class
     seed : int, default = None
         the seed of the random generator
-    classifier : sklearn compatible classifier, default = None
+    estimator : sklearn compatible classifier, default = None
         if None, a RidgeClassifierCV(alphas=np.logspace(-3, 3, 10)) is used.
     n_jobs : int, default -1
         Number of threads to use for the transform.

aeon/classification/shapelet_based/_sast.py

@@ -34,7 +34,7 @@ class SASTClassifier(BaseClassifier):
         the number of reference time series to select per class
     seed : int, default = None
         the seed of the random generator
-    classifier : sklearn compatible classifier, default = None
+    estimator : sklearn compatible classifier, default = None
         if None, a RidgeClassifierCV(alphas=np.logspace(-3, 3, 10)) is used.
     n_jobs : int, default -1
         Number of threads to use for the transform.