I have two data sources. Temperature and Humidity from a single system. I need to analyze whether the system nature is anomalous nor not?
I’d highly recommend looking into the hotgym example:
This example has 2 fields, a timestamp and an energy consumption float. So you could emulate the ‘consumption’ encoder settings with your temperature & humidity. If you don’t have a timestamp to include, you can just drop any ‘timestamp’ values from the ‘encoders’ dict.
Yes , thank you.
I did the separate encoding for both temperature and humidity. Then concatenate and pass to spatial pooler. What’s next?
In this case what will the prediction function predicts?
How anomaly is computed? Is it for the entire data temperature and humidity? How to modify it?. Please support.
Then that SP outputs goes to TM, as standard.
If it’s a TemporalAnomaly model the predictions are of system state at next time step, in the SDR form – not in the raw data form. In order to convert from SDR predicted states back to raw data value a Classifier is used.
There’s one anomaly score per model, so if its a multivariate model the anomaly score represents a mix of all variables involved.
One thing I do sometimes is have separate models for each variable, to get a separate anomaly for each.