Is there a strategy to "hide" noisy measurements?

Is there a strategy to “hide” noisy measurements?

I’ve been thinking about how I can “hide” noisy measurements in Dashboard graphs.
I don’t want to use approaches like grouping by mean in dashborad, so as not to impact the value of the correct measurements.

The best way I could think of (but it’s not currently available) was to apply some machine learning approach to the widget configuration, similar to what is done with the “Transform” option. A “Filter” option with the possibility of implementing approaches to filter (hide) noisy data (e.g. Kalman filter…).

Hello, @alvarolb and @jaimebs

Would this implementation by Influxdb be possible as described in the issue?

We need an answer (yes or no) so we can think about alternatives and possibilities.

If so, what would be the deadline for making it available?
If it is not possible to implement, we will research other possibilities. But we just need a response from the Thinger developers.


Hi @George_Santiago,

In the upcoming release, we’re excited to introduce new features to enhance the dashboard experience, one of which is particularly relevant to your query. We’re integrating what we call ‘Dashboard Functions.’ These are user-defined functions that process each data point before it’s displayed on a widget:

These functions can be individually applied to different data sources, giving you the flexibility to tailor data processing as needed:

Regarding your concern about filtering noisy measurements, these functions open up a range of possibilities. You can implement various heuristic approaches, including anomaly detection in time-series data. However, it’s important to consider the inherent challenges in real-time anomaly detection. For example, distinguishing a genuine anomaly from a sudden but legitimate pattern change can be complex without context. In your example it is “easy” as you see all your data at once, but what happens with future data points you are going to receive?

One approach could be modeling your sensor data and applying techniques like the Kalman Filter. By comparing sensor readings against model estimations, you could potentially identify discrepancies that signify anomalies. However, be mindful that this method requires fine-tuning to your specific use case and may still be susceptible to errors.

I hope this helps, and we’re looking forward to hearing your feedback on these new features!

1 Like