I understand tumbling window is set for an interval and the event's don't overlap and expires at the set time interval. Now both hopping and sliding windows overlap and in case of a hopping window we have a hop interval along with the normal window interval and sliding window has a sliding interval other than the window interval. I looked at this link on Hopping Window and i understand it but the sliding interval seems to be same as well. Could someone help me in explaining the difference.
Asked
Active
Viewed 5,885 times
7
-
I am also confused by this - but the following link aims to differentiate between them by stating that sliding window is a fixed size window which slides over time unlike hopping window. It's still a bit confusing bt may be it answers your question. https://kafka.apache.org/0110/documentation/streams/developer-guide#streams_dsl_windowing – Biplob Biswas Jul 25 '17 at 09:39
-
@BiplobBiswas Thanks I will take a look on it. – Vignesh I Jul 25 '17 at 09:46
1 Answers
9
Microsoft has a good explanation (link). Basically, a hopping window always advances by a specific time interval from the start of the time series. A sliding window only advances when there is data in the time series to evaluate.
Dataset:
- Time T: 18
- T+1: 12
- T+7: 20
- T+11: 15
- T+16: 19
- T+27: 107
Hopping: Starting at time T, sum values for 10 second window, advance by 5 seconds
- T-T10: 50 (18+12+20)
- T5-T15: 35 (20+15)
- T10-T20: 24 (15+19)
- T15-T25: 19 (19)
- T20-T30: 107
Sliding: Starting at time T, sum values for 10 second windows
- T: 50 (18+12+20)
- T1: 47 (12+20+15)
- T7: 44 (20+15+19)
- T11: 24 (15+19)
- T16: 19 (19)
- T27: 107 (107)
The Hopping window is more schedule based, producing output on every period for exactly that period. The Streaming window is more event based, producing output only when data is present.
C Sayers
- 131
- 1
- 8