[BUG] Temporal imprortance curves for DrCIF #3140
Description
Describe the issue linked to the documentation
the classifier DrCIF has a function temporal_importance_curves which should return the importance of each time point for classification
from aeon.datasets import load_classification
from aeon.classification.interval_based import DrCIFClassifier
X, y = load_classification("GunPoint", split="TRAIN")
print("Fitting DrCIFClassifier on GunPoint")
dr = DrCIFClassifier()
dr.fit(X,y)
print("Fit complete DrCIFClassifier")
curve = dr.temporal_importance_curves()
print(curve)The function is inherited However, from BaseIntervalForest, but the above code gives
Fitting DrCIFClassifier on GunPoint
Fit complete DrCIFClassifier
Traceback (most recent call last):
File "C:\Code\tsml-eval\local\debug.py", line 26, in <module>
curve = dr.temporal_importance_curves()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Code\tsml-eval\venv\Lib\site-packages\aeon\base\_estimators\interval_based\base_interval_forest.py", line 1162, in temporal_importance_curves
raise ValueError(
ValueError: base_estimator for temporal importance curves must be ContinuousIntervalTree.But DrCif now uses the scikit DecisionTreeClassifier rather than the bespoke interval tree. Solution is either to remove functionality from DrCIF or somehow implement this desirable feature around the DecisionTreeClassifier. @MatthewMiddlehurst any thoughts?