This is really adding a feature that describes how the delay distribution should be interpreted. That is "what data frequency does the delay distribution correspond to". This is mainly important for unequally spaced observations. But even if the input is evenly spaced, CV results in uneven spacing. The result is that automatically determining the relationship between the delay distribution and x is potentially dangerous.
This fix adds an argument to describe that relationship explicitly. It will attempt to determine it automatically, and fail if it can't (or if downstream issues arise) with meaningful error messages.
Features were added to CV to allow for standard (irregular) splits. This tends to result in better CV error curves.
A subtle change was made in the calculation of the delay distribution so that the first element will be the lag 0 delay. Potentially useful in other contexts, and more mathematically useful.
This is really adding a feature that describes how the delay distribution should be interpreted. That is "what data frequency does the delay distribution correspond to". This is mainly important for unequally spaced observations. But even if the input is evenly spaced, CV results in uneven spacing. The result is that automatically determining the relationship between the delay distribution and
xis potentially dangerous.