Hi to creators of amazing tool! I'm using time series with seasonality (Day of the week). When seasonality is switched on, seems that days in the config becomes weeks. It could be not a huge deal for describing a training period, but it doesn't work for the detection period. For example, my time-series test config:

seasonality: day_of_week
              time_bucket: 
                period: day
                count: 1
              detection_period:
                period: day
                count: 1
              training_period:
                period: day  #weeks in reality
                count: 8
              detection_delay:
                period: day
                count: 0

And how it is reflected on the chart: The problem is that the value already returned to the expected threshold range, the two last days are green, but the test continues alerting. I think it is because of detection period which is also considered in the weeks, like the training period. I don't want it to be noisy a week after the incident. Is it possible to fix or overcome this behavior somehow?

Hey @Dmitrii Khriashchikov,

seasonality is switched on, seems that days in the config becomes weeks

What we're basically doing when

day_of_week

seasonality is configured is we're comparing Sunday to Sunday, Monday to Monday, etc... Therefore, if a training period of 8 is configured, we'll have to check 8 Sundays back, 8 Mondays back, etc. You get the alert a week after for Jan 1st or Feb 7th ?

Thank you, Daniel. I figured out where I was wrong. Sorry that forgot to close the topic here

What was the problem ?

In some tests, a detection period was 2 days, and in some 1 day. I didn't notice and thought that it was 1 day everywhere.

got it. thanks for updating 🙏