Issue of lookback data in Analytics

[tallrain]

Hi,
I have met an issue of lookback data in Analytics.
Now I have a time series TTF data, and trained a model with lookbacksize of 16.
And I mapped input features with nasaTTF_inputThing, with Value Stream configured.
Then I updated the nasaTTF_inputThing to trigger the Analytics Event.
You can see that all features are updated in the same moment.
But the Analysis Job shows that for the input data, 15 of 16 records are from same row, only last record is from previous row of Value Stream.
My understanding is that the so called lookback (with size 16) data should be coming from the previous 16 records of the Value Stream of nasaTTF_inputThing.
This issue makes time series prediction failed.
Can you please help to check - is there anything I missed or configured wrong?
Thanks a lot!

[cmorfin · Accepted solution]

Hi @tallrain 

The behaviour you see is mostly expected.
Analytics Manager collect the last 16 entry within the timesampling interval, so in your case within the last 16 ms.
Since in most cases you do not have new entries within this time interval, the entries are duplicated.

I recommend reading https://www.ptc.com/en/support/article/CS293493 as it explains how this works.

Kind regards

Christophe

[tallrain]

Thank you Christophe!

Where can I set Data Sampling Frequency?

[cmorfin]

The Data Sampling interval comes from the model.
If the model was created in ThingWorx Analytics, then it is defined at the dataset level.

This is the time interval in the csv dataset file for the Temporal field which needs to match the timesamplinginterval defined in the dataset json file. See https://www.ptc.com/en/support/article/CS258345 for some more info on this.

[tallrain]

Thank you, it's working now!

Turns out I need to do the setting in dataset configuration(csv and json files).