hey all i am surprised from <https docs elementary data com Elementary Community #community-support

hey all, i am surprised. from <training_period>: `...

michael böhm

05/09/2025, 9:04 AM

hey all, i am surprised. from training_period:

Copy code

… includes the training period and detection period.

and indeed, i checked the calculated values, the tests’ training period always includes the detection period 😵 there seems to be no setting to make training exclude detection. 1. how can i make training exclude detection? 2. why does training include detection in the first place?

michael böhm

05/09/2025, 9:06 AM

fyi, tested on v0.18.2

michael böhm

05/09/2025, 9:51 AM

or more general: how can i prevent any tested datapoint to be in its own training set?

michael böhm

05/09/2025, 2:50 PM

to the devs: apart from excluding the datapoint, another trivial improvement would be to use the median instead of the avg when calculating the number of stdevs a.k.a. anomaly score. this way, if the datapoint is anomalous, it likely will not affect the median (unlike the avg).

Noya Offer

05/12/2025, 9:14 AM

Hey Michael, understand the frustration and appreciate the feedback. This was a decision taken a while back, and we're highly aware of the fact that it needs to be changed. We're planning to enable this within the quarter, happy to get contributions if can be relevant for you 🙂 Taking the last comment into consideration as well within our anomaly detection improvement effort. We shared an anomaly detection feedback form - feel free to add any more feedback!

michael böhm

05/12/2025, 9:27 AM

will use the form, thanks!

🙌 1

6 Views

Open in Slack

Previous Next