Can the Anomaly Detection be trained at a sampling rate and opertionalized at a different one?

[nahuel]

Hi community,

Can the Anomaly Detection functionality be trained at a sampling rate and opertionalized at a different sampling rate?

Use case:

Let's say I have a dataset that has 144 records from an asset that has been sampled at sampling rate of 10 minutes (doing the math that is one day of data).

Since the only way to train the model behind the Anomaly Detection functionality is through streaming data, I would have to run a simulation to train the model. Knowing that once the model is trained it will be migrated to the production environment where the sampling rate is 10 minutes, do I have to run a one day long simulation or could I speed up the process by reducing the simulation/training sampling rate?

[nsampat · Accepted solution]

@nahuel ,

Thank you for posting your question to the PTC Community.

Anomaly Detection relies on the timestamps of the data points, not the computer's time. If your pre-collected dataset contains timestamps that have a sampling rate of 10 minutes, you can input those into Anomaly Detection as fast as your able without any negative side-effects. 

You can train via simulation, you would just need to make sure that you're inputting both the timestamp and value of the pre-collected data into the ThingWorx property - not just the value.

Please let me know if this answers your question by marking this comment reply as "Solution Accepted".

Regards,

Neel