TimeframeFeatureCorrelationCalculator cuts the observations within the window at the start and appends them to the end

[gkronber]

The expected behavior is that the calculator determines correlation with lagged features.

          IEnumerable<double> var1 = dataset.GetDoubleValues(variable, indices);
          IEnumerable<double> var2 = dataset.GetDoubleValues(doubleVariableNames[i], indices);

          var valuesInFrame = var1.Take(j);
          var help = var1.Skip(j).ToList();
          help.AddRange(valuesInFrame);
          var1 = help;

          var error = OnlineCalculatorError.None;
          elements[i, j] = calc.Calculate(var1, var2, out error);

I cannot think of a case where it would be correct to just append the first few observations to the end of the time series when calculating lagged correlations. Instead only a subset of all rows (for which both variables are known) should be used for calculating the correlation.

[gkronber]

r14300: fixed TimeframeFeatureCorrelationCalculator

[mkommend]

Reviewed r14300. Thank you for spotting and correcting this bug.

[gkronber]

r14318: merged bug fix r14300 from trunk to stable