Moving Average

This node calculates the moving average of a column. The moving average values are displayed in a new column appended at the end of the table or (if selected) replaces the original columns. For all window based methods (Backward/Center/Forward simple/Gaussian, Harmonic Mean) the cells that do not have a complete window (at the beginning and the end of the table) are filled with Missing Values.

Options

Columns containing Double Values

Select the input column containing double values on which to perform the moving average.

Window Length

The number of samples to include in the moving average window. It has to be an odd number if a center based method was selected. Minimum value: 3 samples. Maximum Value: Time Series length.

Remove original columns

If selected the original columns are replaced with the moving average columns.

Type of Moving Average

Moving Average can be applied with various methods. Here the used formulas for every kind, where v_n is the value in the n-th row of the data table in the selected column and k is the window size.

Methods

Backward simple

Backward_simple_n = 1/k * sum{v_n-(k-1) ... v_n}

Center simple

Center_simple_n = 1/k * sum{v_n-(n-1)/2 ... v_n ... v_n+(n-1)/2}

Forward simple

Forward_simple_n = 1/k * sum{v_n ... v_n+k-1}

Backward Gaussian

Backward_gaussian_n = sum{i = 0 ... k-1} gauss(i,k-1,std_dev)*v_n-i

Center Gaussian

Center_gaussian_n = sum{i = 0 ... k-1} gauss(i,(k-1)/2,std_dev)*v_n+(i-(k-1)/2)

Forward Gaussian

Forward_gaussian_n = sum{i = 0 ... k-1} gauss(i,0,std_dev)*v_n+i

Harmonic Mean Center

The harmonic mean can only be used for strictly positive values.

Center_harmonic_n = n/{sum{i = 0 ... k-1} 1/v_n+(i-(k-1)/2)

Cumulative simple

Cumulative_n= 1/n * sum{v_0 ... v_n-1}

Simple exponential

Simple_exponential_0 = v_0

EMA(v,n) = Simple_exponential_n = alpha*v_n + (1-alpha)*Simple_exponential_n-1

Double exponential

Double_exponential_n = 2 * EMA(v,n) - EMA(EMA(v,n),n)

Triple exponential

Triple_exponential_n = 3 * EMA(v,n) - 3 * EMA(EMA(v,n),n) + EMA(EMA(EMA(v,n),n),n)

Old Exponential

Exponential_n = alpha*v_n + (1-alpha) * Backward_simple_n-1

Appendix: Gaussian

For the Gaussian weighted moving average the individual values are weighted based on the position in the window.

std_dev = (k - 1) / 4

and the weighting:

gauss(i,mean,std_dev) = Math.exp((-0.5) * (i - mean) * (i - mean) / std_dev^2)

Appendix: Exponential

alpha = 2/(k+1)

Views

This node has no views

Workflows

• 01_Data_PreparationKNIME Hub
• 01_Data_PreparationKNIME Hub
• 01_Example_of_Time_Series_FunctionalityKNIME Hub
• 01_Example_of_Time_Series_FunctionalityKNIME Hub
• 01_Guided_VisualizationKNIME Hub
• Show all 147 workflows

Links

• Need help (very new to this field)by ipazin on 2019-03-11

  forum.knime.com/p/55946
• Choice for time series loopingby HansS on 2019-11-21

  forum.knime.com/p/75296

Developers