Need some help setting up a Filter Sensor

Hey everyone,

I have a sensor with values that range from 0.005 to 0.015 that updates every minute or so.

For some reason I cannot figure out or troubleshoot, there are some values, twice a day, that go up from 4 to 30.

This really messes up my graphs.

I was trying to set up a filter sensor like this:

  - platform: filter
    name: "filtered sensor"
    entity_id: sensor.original_sensor
      - filter: outlier
        window_size: 1
        radius: 0.01

I have tried different window_size and radius values, but I cant figure out which ones to use, I cannot quite figure out the documentation details.
Right now with the above values i get a constant 0 with some peaks around 1-2.

How can I create a filtered sensor that will ignore any value larger than say 0.2?

Thanks in advance.

As far as I understand it, the window size is how many samples to use for a rolling average. The new value must be within +/- radius of this average to be accepted as valid.

I have found strange things happen after a restart though. If the first sample is a bad one it will make it into the average then no real readings are accepted. :roll_eyes:

As you have a well defined range of 0.005 to 0.015 perhaps you would be better off with a range filter?

      - filter: range
        lower_bound: 0.001
        upper_bound: 0.02

I never even noticed the range filter.
My fault for not going through the whole documentation page :stuck_out_tongue: It looks like exactly what i need.
Updated the code and rebooting now.
Let’s see

I did a test run with upper 0.1 and lower 0.
It seems to be working but it is rounding the results to two decimals. I am getting a constant 0.01.
I’ll let it run a couple of hours, maybe it will correct itself after it has some more samples.

There is a precision option but it seems like its referring to another filter, so I dont know if it can be used.

Thanks for the input @tom_l

I think precision should work for all the filters, lowpass is just mentioned in the docs an example.