https://github.com/nixtla logo
m

marah othman

06/14/2023, 3:01 PM
hello , i have question about detect seasonality in time series is there any method to know if i have seasonality in my data without plotting and graph ? i mean somthing static thank you
f

fede (nixtla) (they/them)

06/14/2023, 3:10 PM
Hi @marah othman! Here’s a discussion on the topic. In particular, the image attached. You can calculate the strength of the seasonality using the tsfeatures library and computing the
stl_features
function.
m

marah othman

06/14/2023, 11:33 PM
are you using in neuralforcast library this feature to train the models ?
b

Bartosz Bohaterewicz

06/20/2023, 8:37 AM
ACF, sort by descending and look at the second coefficient
m

marah othman

06/20/2023, 8:45 AM
the second coefficient you mean the second lag ? and could explain why the second coefficient ?
b

Bartosz Bohaterewicz

06/20/2023, 10:00 AM
because ACF starts at lag of 0 (first) which will always equal to 1 due to the fact that you are correlating TS with itself. In addition you can look at the index higher than 6 in case of daily data (weekly) and apply custom r-value threshold
m

marah othman

06/20/2023, 10:11 AM
mu data is hourly so you told me to see the second and now you told me see the 6 so i am little confuse
b

Bartosz Bohaterewicz

06/20/2023, 10:32 AM
what kind of seasonal lag are you expecting to see in your dataset?
m

marah othman

06/20/2023, 12:31 PM
how can determine this ?
my seasonality each 6 days or even every day i am really not very sure because it different when i take the frequency for 1 hour or for one day or five day
2 Views