Standard (interpolated) non-parametric cluster trajectory estimation through lcModelPartition and lcModelWeightedPartition

[niekdt]

Many methods and models (e.g., lcModelPartition, lcMethodStratify) depend on a non-parametric estimate of the cluster trajectories based on the original data. Having a standard exported method clusterTrajectories() with signature data.frame for computing such cluster trajectories would make it more transparent to users how these centers are generated, and which additional arguments are available (which could be provided through the method specification).
Moreover, it enables users to use the center computation in their own models.

• Define clusterTrajectories(data.frame, assignments, center = meanNA, at = numeric())
• Move functionality of computeCenterClusterTrajectories() to clusterTrajectories(data.frame)
• Update methods and models using any type of center specification
• Extend clusterTrajectories(data.frame) to interpolate between measurements.

[niekdt]

On second thought, this functionality is better moved to lcModelPartition and lcModelWeightedPartition. Otherwise, the clusterTrajectories(data.frame) function would need two handle two cases:

1. Hard partition, with a center function that accepts a data vector
2. Fuzzy partition, with a center function accepting the full data vector along with a weight argument

• Move hard cluster trajectory computation code inside clusterTrajectories(lcModelPartition)
• Extend clusterTrajectories(lcModelPartition) to interpolate between measurements.
• Move fuzzy cluster trajectory computation code inside clusterTrajectories(lcModelWeightedPartition)
• Extend clusterTrajectories(lcModelWeightedPartition) to interpolate between measurements.
• Update methods and models using any type of center specification
• Center specification as a method argument, used by default when calling clusterTrajectories() of the respective model.