GroupNormalizer gives odd results in pytorch-forecasting

data
1

Hello everyone,

I am currently playing with some dummy data to perform deep learning tasks. I was checking the normalization of my data, which seems wrong.
Here is how my data is defined:

num_rows = 100
df = pd.DataFrame({
    "time": range(num_rows),
    "temperature": [100*(-1)**n for n in range(num_rows)],
    "group": [1]*num_rows
})

Clearly, the temperature has a mean of 0 and a standard deviation (scale) of 100, however if I run the following

gn = GroupNormalizer(groups=["group"])
gn.fit(df["temperature"], df)
print(gn.norm_)

I end up with the following results:

       center       scale
group                    
1         0.0  100.504758

That’s why I was wondering what is going on.
Is the error coming from numerical approximations ? (but the error seems large in comparison of the number of rows)
Is it coming from the way GroupNormalizer computes the normalization ?
Am I missing something ?

Thanks for reading!