The aim of data condensation is to create a set of observations that has significant smaller count than the observation set of the source feature but is still a reasonable representation of the real world phenomenon. Condensing data might be a good choice to get a reasonable sized set of observations that can be manually examined, presented graphically, or that can be further processed with reasonable performance.

The data condensation: simple calculation type provides very simple condensation algorithms that do not find a reasonable representation of the real world phenomenon in all cases, but are fast in terms of processing and prove to be useful if you can make certain assumptions on the characteristics of your source data.

Table 1: Calculation brief

Table 2: Input pins

Table 3: Configuration settings

If the calculation is the final calculation of the algorithm the used classifications of the source features or calculations has to be the used classification in the domain of the calculated feature. If this is not the case the calculation will fail.

Each n-th Algorithm

This algorithm simply takes each n-th observation of the source feature or calculation. Figure 1 shows a formal specification of the algorithm.

The configuration setting n-th influences the condensation ratio defined as NumberOfGeneratedObservations / NumberOfOriginalObservations. Higher values for n-th produce smaller condensation ratios.  If set to true the configuration setting include last observation will always add the last observation of the source feature or calculation to the output.

Figure 2 shows an example output of the Each n-th algorithm.

The each n-th algorithm is useful when dealing with observations recorded at a high and regular sampling rate but with little changes in the observed property values and if the exact time when a property value changed is not important.

Figure 1: Selection of output observations by the Each n-th Algorithm.

Figure 2: Example of Each n-th algorithm

One per interval        

This algorithm will choose only one observation per interval. The intervals are defined by the Start timestamp and Time interval configuration setting  by using the algorithm shown in figure 3. Figure 4 shows which observation is picked from all the observations found in an interval.

This algorithm is useful if you have observations at a high and irregular sampling rate with little changes in the observed property values and if the exact time when a property value changed is not important.

Figure 3: Specification of the intervals for the on per interval algorithm

Figure 3: Picking one observation per interval

Examples for picking an observation from an interval:

•If the interval contains observations with sampling timestamps {t1, t2, t3, t4, t5}, t3 will be chosen.

•If the interval contains observations with sampling timestamps {t1, t2, t3, t4}, t2 will be chosen.

•If the interval contains observations with sampling timestamps {t1, t2}, t1 will be chosen.

•If the interval contains observations with sampling timestamps {t1}, t1 will be chosen.