Training Period

In a cross-validation test, such as the one you can perform in Windographer's LTA (Long Term Adjustments) window, the training period is the portion of the available data used to train the algorithm. If you are testing the linear least squares algorithm, for example, you would use the data in the training period to calculate the line of best fit.

The testing period is the portion of the available data used to test the algorithm.

Often the training period and the testing period are complementary subsets of the available data, meaning that the training and testing periods do not overlap: each time step falls in one or the other, but not both. The following data coverage diagram shows an example of complementary training and testing subsets:

In the above example, the available data (solid grey bar) covers all of March 2014. The training period, in orange, covers part of the first day of March, then part of the second and third days, then the fifth day, then the seventh day, and so on. The testing period, in blue, covers all the segments missing from the training period. No overlap occurs between the training and testing periods in this example.

The diagram below shows a different example, in which the training and testing periods cover all of March 2014: