Filter

A single filter definition

Table of content

• Properties
  • Measure
  • Mean Methods
    • Average (Arithmetic mean)
    • Median
    • Mode (Shortest or Longest)
    • Central
    • Fixed

The Mean Filter node is useful to filter out statistical aberrations.

Properties

Keep in mind that calculations are relative to the mean value.

Example

If using Relative measure and Average mean method, the mean value will be the average of all values.
The Exclude Below value is substracted from that average to find the minimum acceptable value (in relative terms), while the Exclude Above value is added to that average to find the maximum acceptable value.

In other words, if the average relative mean is 0.5, using Exclude Below = 0.1 and Exclude Above = 0.1, values < 0.4 and > 0.6 will be excluded.

Measure

Mean Methods

The mean method is used to find the reference threshold value that will be used by Exclude Below and Exclude Above.
Below are an explanation on how each method works.

For the example, let’s say we’re working on a dataset with 10 points that have a float attribute with the following values:

Absolute values = {10, 20, 30, 40, 50, 50, 55, 60, 1200, 500}
Relative values = {0.0083, 0.016, 0.025, 0.03, 0.041, 0.041, 0.045, 0.05, 1, 0.41}

• Average
• Median
• Mode
• Central
• Fixed

Average (Arithmetic mean)

Average mean is, well, the averaged value of all the values.

Absolute Average Mean = 201.5
Relative Average Mean = 0.125

Median

Median uses the median value of all the available lengths, sorted.

Sorted Absolute values = [10, 20, 30, 40, 50, 50, 55, 60, 500, 1200]
Absolute Median Mean = 50 // [.., 50, 50, ...]

Sorted Relative values = [0.0083, 0.016, 0.025, 0.03, 0.041, 0.041, 0.045, 0.05, 0.41, 1]
Relative Median Mean = 0.041 // [.., 0.041, 0.041, ...]

Mode (Shortest or Longest)

See Mode (statitics) on Wikipedia.

If there are concurrent mode values (multiple buckets containing the same amount of values), the selected mode variant allows you to select either the mode with the smalles values, or the one with the largest out of the available conflicting modes.

Absolute Mode Mean = 50 // [50, 50] is the largest bucket of equal values
Absolute Mode Mean = 0.041 // [0.041, 0.041] is the largest bucket of equal values

In this is scenario, there is no conflicting frequencies.

The Mode Tolerance property is used to fill frequency buckets based on equality with already sampled values.

Central

Central uses the middle value between the smallest and largest values as mean.

Absolute Central Mean = 605 // 10 + (1200-10)/2
Relative Central Mean = 0.504 // 0.0083 + (1-0.0083)/2

Fixed

Fixed mean is basically user-defined mean.
This is the way to go if you don’t care about statistics or if you have a consistent, metrics-driven setup.

This is the least flexible but most performant approach as there is no need to compute any statistics prior to pruning.