Splunk Predict App: How to use SPL to forecast

[parveen77]

New to Splunk and trying to follow this example here:
https://discoveredintelligence.ca/forecasting-time-series-data-using-splunk-machine-learning-toolkit... using my data

time value
2019/07/02 22:35:00 1248.2
2019/07/02 06:15:00 1266.05
2019/07/02 06:35:00 1274.17
2019/07/02 06:45:00 1277.12
2019/07/02 19:50:00 1277.8
2019/07/02 23:40:00 1279.41
2019/07/02 16:15:00 1285.74
2019/07/02 20:20:00 1286.51
2019/07/02 06:25:00 1287.84
2019/07/02 20:10:00 1288.02
2019/07/02 06:30:00 1289.44
2019/07/02 22:20:00 1290.35
2019/07/02 06:05:00 1292.87

But I get this error:

External search command 'predict' returned error code 1. . 

I checked answers to similar issue and recommendation was to create a timechart.
The example I have been following and running in Splunk has not done so and the example runs fine for me, but with my data I get error.

Any pointers would be super helpful.

Thanks

[Sukisen1981 · Accepted solution]

hi @parveen77
The unit for the span specified with the timechart command must be seconds or higher. The predict command cannot accept subseconds as an input when it calculates the period.
So if i do this

 index="_audit"
    | stats count(action) by _time 
    | rename count(action) as count 
    |  predict count

i get the error code -1
but if i just put a span of 1 min and use the same code with the bin command added

index="_audit"
| bin span=1min _time
| stats count(action) by _time 
| rename count(action) as count 
|  predict count

I receive the results.
Timechart is not mandatory and you can run both these code snaps as it is, since i ran this on the delivered _audit index for the last 24 hrs.
If you use timechart you need to use span (min 1 sec), if you use something else like stats or a table you have to take care of the time buckets using the bin command

[Sukisen1981]

hi @parveen77
Please accept the answer if it significantly helped resolve your issue or let us know if there are any more issues