Nima Sarajpoor

Yeah...I also tried many many different ways to resolve this. I have not been successful so far. ---- FYI: > (The indices 2 and 12 are in fact the start...

> What happens when you `print (1.0 / ref_std[2])` and `print(1.0 / comp_std[2])`? ``` >>> 1.0 / ref_std[2] 971803.3300204428 >>> 1.0 / comp_std[2] 971802.1137688879 ``` In case that matters: Since...

I think this should give you the `T` you are looking for. ``` def identical_cases_with_scale(): m = 3 zone = int(np.ceil(m / 4)) seed = 27343 np.random.seed(seed) identical = np.random.rand(8)...

> This might not work for the multi-dimensional case but I hope you can get my point After I saw your suggestion, I tried it by calculating an upper bound...

@seanlaw Btw, I would like to mention that the current solution, `i.e. checking std and re-calculating its exact value` may not be an efficient solution if ALL VALUES of time...

> We also have to ask ourselves how likely are we to come across strange time series data like the contrived examples that we have? Correct! I think the current...

**Update** > > This might not work for the multi-dimensional case but I hope you can get my point > > After I saw your suggestion, I tried it by...

> Can you help me understand what that means and if there is anything we could do? So I did `fastmath=False` to remove its impact on imprecision if there is...

@seanlaw Very interesting. I have not read the paper yet. This is just what I think: I think we should simply replace a constant subsequence with all zeros (after z-normalization)....

Another thing that just came to my mind and I thought it might be worth sharing: The domain of z-normalized version of a non-constant subsequence with length `m` is `R^{m}`...