How to get the trend line as my trend line is not showing
Hi Kokai! In the data output of the Decompose Signal node you have a column called Trend Component. The Line Plot node with the rowID on xaxis and the Trend Component column on the yaxis gives you the trend line, like this:
Oh ok, so i understood i can get my trend line by using the trend componenet in the line plot. But why is my trend line looking like this in decompose signal but when I tried it in line plot it look like this?
Sorry but I am very new to this software and have some more enquires as the knime didn’t have much information for me. All I see and learn from information in the KNIME for the decompose signal is based on the selected column I wanted, they will get the seasonality and trend then they start to remove both data from the signal and get the residual value. These are some enquires I have:

How do I know what value should I be putting for lag step or max lag?

Should I be also change my correlation cutoff? As I have seen one forum where it says to lower down it to 0.2 so that it can get a better outcome.

Based on the below image, is the max lags consider to be number of day?

How can i understand the ACF chart in the interactive view?

What the meaning if the black line go out of the red boarder?

Should I be looking at the seasonality 1 or 2 for ACF? As I realised there not much a different in both
Hi Kokai,
you’re welcome to join us with these questions. Regarding your first question, the interactive view of the Decompose Signal component doesn’t show the trend line itself but the signal values after the trend has been removed (the line plot in the second row and first column)
My answers to the numbered questions below:
 Max lag should be at least as high as you expect the length of the seasonal cycle to be. So, for example, if you expect weekly seasonality in daily data, then the max lag should be at least 7. Normally, you want to consider all values up to the max lag and therefore leave the lag step to 1. And in addition, it is easier to recognize the seasonal pattern if you use max lag greater than the length of the seasonal pattern. For daily data with weekly seasonality you can set it to 50 to see the weekly pattern repeating at lag 7, 14, 21, and so on.
 Correlation cutoff should only be changed if the component isn’t able to decompose a seasonal component from the data with the default value 0.5. Lowering the cutoff value enables extracting a seasonal component where the correlation between the seasonally lagged values is weaker.
 If you have daily data, then the max lag is the number of days
 The ACF chart shows the autocorrelation values at all lags. The first peak in the curve shows the maximum autocorrelation value which determines at which lag the 1 seasonal component occurs given that it is above the cutoff value. The next peaks occur at multiplications of this lag.
 If the black line goes outside the red borders, then the nonzero autocorrelation value is statistically significant. In your case this happens only once so you might want to proof whether it’s repeating by increasing the lag value to a multiplication of 7. It might also be some noise in the data that causes this.
 According to your view there is 1st but not 2nd seasonality. The bottom left line plot shows the residual of the signal after the decomposition. The bottom ACF plots are the same because the last step, removing 2nd seasonality, didn’t take place.
I hope this helps you!
Best,
Maarit
Hi There,
I m learning time serie and i have execute the decompose signal.
I m a little confused with all this lines How can i know what line is relevent here? And how can i read this graphik?
Thanks a lot.
Marija
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.