Tony Bagnall

I think I have figured out the cause, I'll attempt the fix then make more significant changes in a second PR

see #1513

hi, thanks @CodeLionX I'm fine with that solution, but would also like to know why it does it, bet its a numba thing :)

@aadya940 I think your solution is the simplest and most practical. This is assuming there is no logic in the operation of the distance function, but tbh LCSS is not...

I cant actually reproduce this with the above test, but will put in the fix nevertheless, cant hurt

having looked into #1498 and found a bug that numba was hiding, it might be good too look at these failures in more detail, so will list here - [...

so I think both fails are caused by catch22, and by a tedious process of elimination I have isolated the problem to this function. Error happens when you comment out...

ok, I have narrowed this down to this operation if nsum2 > 0: nsum2 /= len(X) - 1 nsum += nsum2 * np.log(nsum2) basically with numba turned on, 19/99 Numba...

ah I think I have found it. ```python indicies = np.argsort(X) ``` give different results for the same X with and without numba. If the one in a different order...

and the two values are the same print(x1[50], x1[51]) print(x1[36], x1[37]) gives print(x1[50], x1[51]) print(x1[36], x1[37])