I am monitoring sensor data with ML. Each time the sensor is replaced, the values jumps, that's something I am OK with. But after the jump, the model bounds are getting wider. I don't want this, and I want to restart from scratch.
My job is a partition_field_name one, means that the job is split by sensor. So I want to reset only 1 model, not the others
thank's for your help
You cannot reset the model for an individual partition - only the job as a whole
Just so you know - the model should relatively quickly adapt to the new sensor's behavior. ML is designed for this so that there's less burden on model maintenance.
By the way, do you really only get sensor information once per day as your screenshot suggests?
Around 1 per day, yes. Is it a problem?
Nope, not a problem. Was just curious as to how "real-time" your data actually was.
Anyway, I do not find that the model adapts very quickly. Here is an other example. As you can see it took like for the model to 15 days to drop down and the accuracy is very low for seveal months from this moment.
What version are you using?
@dao - What version are you using?
Sorry, I have 2 clusters, one in 6.2.2 and the other in 6.5.1
Ok thanks - i ask because in v6.4 - a specific feature was put in to allow the model to adjust more quickly after a state change: https://www.elastic.co/blog/designing-for-change-in-elastic-machine-learning
However, even so, with only daily samples, adjusting to change still may seem like a long time (in terms of time), but in terms of the number of samples, it's actually not that much.
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.