Line attribute "smoothing" when using "shape: spline" has no effect

[patfelst]

To smoth line graphs, I set shape to spline. This works ok, but I wanted to tune the amount of smoothing as documented here, however changing the smoothing values seems to have no effect on the line.

Without spline

With spline and any value of smoothing

type: custom:plotly-graph
entities:
  - entity: climate.ac_front_lounge
    line:
      color: red
      dash: solid
      shape: spline
      smoothing: 0.5
      width: 1
    attribute: current_temperature
hours_to_show: 24
refresh_interval: 10
period: auto
title: Air con temp sensors

Am I doing something wrong, or is this a bug? Thanks for this great HACS add on.

[dbuezas]

Hi,
the smoothness of the spline only changes what happens between the data points, bot at the data points. Since you have so many data points in the visible range (24 hours), the spline is not noticeable.
Try the moving average filter instead!

[patfelst]

Thanks yes I see what's happening if I reduce x-axis to 2 hours. So I've put in an exponential moving average with alpha of about 0.4 and in combination with spline smoothing of 0.6 can achieve what I want - thank you!

[dbuezas]

screenshot! :)

[dbuezas]

make sure to try the sliding window and try the "extend" parameter, it sometimes gives better results

[patfelst]

Ok so I did some experimenting - and sorry for the long post! Spline is always good (for me) I like the smooth look. Like you said, the smoothing amount really only matters when the number of x-axis points is small (when you zoom right in), so I'm just leaving spline smoothing at 1.0.

Small number of x-axis points, raw (no spline, no filter).

As above with spline + smoothing=0.1 (but no other filter)

As above with spline + smoothing=0.5 (no filter)

Moving on to zoomed out on the x-axis, where spline smoothing makes no visible difference, but applying a filter instead. The Sliding Window Moving Average (SWMA) is better (for me) than the Exponential Moving Average (EMA). The EMA really changes the shape of the curve, where as sliding window retains the shape but smooths out the spikes. This is better demonstrated the my home energy / solar meter.

All of these screen shots use a spline. Raw data (no filter):

EMA alpha=0.5

EMA alpha=0.1

EMA alpha=0.03 (very heavy filter, max values don't have time to reach their true values)

Sliding Window Moving Average (SWMA) window-size=2 (again, all with spline turned on)

SWMA window-size=6

SWMA window-size=20

As you can see in the last image, there is quite heavy filtering, but unlike the EMA, the SWMA retains the peak values, but makes the curve look cleaner.

[dbuezas]

Very nice post!
For completion: Sliding window averaging has two issues:

• Data lags behind
• You lose the data of the first N points, where N is the window size.

To mitigate that, I implemented two techniques:

• Center, which cancels the lag by shifting the data, so for each timestamp the window goes right and left equally
• Extend, which dynamically reduces the window size so the plot can reach the extremes. This means the smoothing will be less effective on the sides.

I think you've enabled both, right?

I will implement trendlines too, which will add other more advanced ways of filtering data, even allowing forecasting. If you like the topic, make sure to subscribe/watch to #187. Feel free to drop ideas and feedback there too

[patfelst]

yes I enabled both. Thanks for the heads up re trendlines!