Bayesian Sensor not updating (always off and "probability" 0 no matter how the inputs change)

Configuration Frontend
1

Hello,

I have several presence sensors (Aqara FP1E) and the iBeacon signal of my Android phone (using the companion app as a transmitter and my HA host, which is in the same room as the presence sensors, as the receiver) and now want to combine all of these with a Bayesian sensor to compensate false-negatives of the presence sensors (while I am sleeping) and false-positives of the BLE sensor (when I leave my phone in the room).
I currently have the following YAML:

binary_sensor:
- platform: bayesian
  name: "Anwesenheit Bayesian"
  prior: 0.5
  probability_threshold: 0.8
  unique_id: 9872abdc-8ca5-4d66-801d-4685f93126b7
  device_class: occupancy
  observations:
   - platform: template
     value_template: >
       {{states("binary_sensor.fpe1e_1_presence")=="on"}}
     prob_given_true: 0.9
     prob_given_false: 0.5
   - platform: template
     value_template: >
       {{states("binary_sensor.fpe1e_2_presence")=="on"}}
     prob_given_true: 0.9
     prob_given_false: 0.5
   - platform: template
     value_template: >
       {{states("binary_sensor.fpe1e_3_presence")=="on"}}
     prob_given_true: 0.9
     prob_given_false: 0.5
   - platform: template
     value_template: >
       {{states("device_tracker.handy_aaron_ble")=="home" and states("sensor.handy_aaron_distance")|float(11) < 8}}
     prob_given_true: 0.7
     prob_given_false: 0.4

However, the binary sensor stays ‘off’ all the time even if all 3 FP1Es are triggered and my phone is within BLE range, the probability atteibute also stays at 0 all the time.
What could be the cause of this issue?

Best regards
Aaron

2

Not sure why you’re using templates. It may not be the cause of the problem, but it complicates things. Why not…

    - platform: "state"
      entity_id: "binary_sensor.fpe1e_1_presence"
      to_state: "on"
3

I had it like that in the beginning but as it didn’t work I changed it. Now I changed it back to what you suggested and it still doesn’t work…

4

I’ve been having a look at my Bayesian sensors. It’s a while since I set them up, but I notice that in all of your observations prob_given_true and prob_given_false added together give a number > 1. In mine the sum of the two is always <= 1. Which sort of makes sense intuitively.

    # Phone in kitchen -------------------------------------------------------------

    - platform: "state"
      entity_id: "sensor.nokia_area"
      to_state: "Kitchen"

      # Phone often gets left elsewhere

      prob_given_true: 0.33
      prob_given_false: 0.02

Have you looked at this post? It’s quite old, but it provides spreadsheets for working out the observations.