Splunk predict command period vs future_timespan?

[kdimaria]

I am wondering if anyone has an explanation of exactly what period is and what future_timespan is? I already read the document http://docs.splunk.com/Documentation/SplunkCloud/6.6.0/SearchReference/Predict which talks about both of the parameters but I am still really confused on what exactly they do and would like for someone to explain them to me in their own words. Thank you!

[jkat54 · Accepted solution]

Period is the data that is crunched in order to determine the prediction.

Lets say i have a single event every day at 1pm with a field that contains a number. The value of that field has decreased by 1 every day for the last 7 days. If i specify a period of 7, the algorithm would see that over the last 7 days, the number has decreased by 1 every day... and therefore is likely to continue decreasing by 1 with very little margin for error in the prediction...

If however the event was up by 10 every day for days 1-3, and down by one every day from days 4-10, and I specify a period of 10... then the algorithm is going to give a wider margin of predictions.

future_timespan is how far into the future to predict results. If you specify a future_timespan of 50 with the previous example, since the data comes in once per day, the predict command would produce 50 days of future predicted data points.

[kdimaria]

Thank you I think I finally get it now

[niketn]

@kdimaria, period and future_timespan arguments are different for sure.

In order to improve prediction you can add period argument with data points after which your data pattern repeats. For example if you have a timechart with span=1d (1 day) and your weekly trends are similar i.e. Every Moday your events rise and every Thursday is your Peak, your events start declining from Friday and Sunday is no/minimal traffic. Then you would define 7 as your period.

| timechart span=1d count as Traffic
| predict algorithm=LLP period=7

The future_timespan argument tells predict command how many future buckets to predict based on your time span selected. i.e. if you have set it to 5, and timechart span=1d it will predict upcoming 5 days.

| timechart span=1d count as Traffic
| predict algorithm=LLP period=7 future_timespan=5

Please let us know if this is what you required or something else?

[kdimaria]

Thank you! I think I understand now. the period was just very confusing.