Tony Bagnall

basically if they are equal it seems that numba call puts the lower index first, normal call the higher one. So short of implementing argsort, this does not seem solvable....

simpler example ```python # @njit(fastmath=True, cache=True) def argsort_normal(X): return np.argsort(X) @njit(fastmath=True, cache=True) def argsort_numba(X): return np.argsort(X) X_train, _ = load_basic_motions(split="train") i =2 j=0 x1 = X_train[i][j] indx1 = argsort_normal(x1) indx2...

closing this, replaced by more specific bug #1646

closing this, testing now in hand

its madness

I think we are in a better place now @lmmentel can we close this?

agreed, this is fixed

yes. its fixed

sorry @baraline I didnt see you comment! I have done a simple comparison using the PeriodogramTransformer ```python from aeon.transformations.collection import PeriodogramTransformer import time import numpy as np p1 = PeriodogramTransformer(use_pyfftw=True)...

y axis is time in seconds