traj 3.1.0

• Changed the eigensolver from eigen to eigs_sym (package RSpectra) to speed up trajClusters().
• Fixed a small bug that occurred in trajMeasures when ID=FALSE and some trajectories have less than 3 observations.
• Made substantial changes to the trajdata data set.
• Fixed a mistake in the trajReduced() function that caused it to output the wrong number of clusters in certain situations.
• Fixed labeling of the columns in the fuzzy.partition component of trajClusters().
• In the scatterplots() function, the title of the plots is more informative.
• To facilitate interpretation, the summary() function, as it applies to the output of the trajClusters() function, now includes a top_p argument which identifies, for each cluster, the top_p measures which are most extreme.
• The plot() function now additionally displays the cluster medians of each standardized measure.

traj 3.0.1

• In the plot function, setting the argument sample.size to NULL plots all the trajectories on the same graph (in a random order).
• In the trajClusters function, setting the argument subset.n to a numerical integer while nclusters is set to NULL makes it so the optimal number of clusters is determined using a random sample of the data of size subset.n. Assuming subset.n is large enough that the random sample is representative of the data, this would speed up the process of identifying the optimal number of clusters.
• In scatterplots, the legend now appear outside the scatter plots, for improved visibility.
• In the scatterplots function, we added an argument which.scatter allowing to plot only a subset of all the available scatter plots.
• In the scatterplots function, we added an argument N allowing to plot a random sample of size N, while preserving the groups’ relative sizes. Assuming N is large enough that the sample is representative of the data, this would speed up the plotting process.

traj 3.0.0

• Substantial modifications to the measures.
• The clustering step now relies on a version of the Spectral Clustering algorithm.
• The main function are trajMeasures (computes the measures) and trajClusters (finds the clusters), with trajReduce (finds a representative subset of measures) being accessory.
• The plotting functions that trajClusters can be passed into are now plot, scatterplots and CVIplot.
• The trajClusters function responsible of finding the clusters has a logical argument that allows to choose between soft and hard clustering.

traj 2.2.1

• Added a new measure, “m5: slope of linear model”.
• Spit the “plot” function into three: plot, plot.scatter and plot.crit.
• Improved the presentation of the scatter plots.
• Made Step1Measures more lenient with how the input data is formatted.

traj 2.2.0

• Added k-medoids as the default clustering algorithm.
• Added the Calinski-Harabasz index as the default criterion for determining the optimal number of clusters.

traj 2.1.0

• Makes substantial changes to the list of measures.
• In trajdata, the group of size 30 is made up of quadratic (instead of linear) curves.
• Introduces the vignette “Using the traj package”.

traj 2.0.1

• Fixes minor bugs and improves presentation.

traj 2.0.0

• Changes the way the measures are computed in order to be better suited to missing values and unequally spaced observation times.
• Allows for better control over the choice of a “midpoint”.
• Implements a capping procedure to automatically handle outliers. Measures of the form 0/0 are set to 1.
• Fixes an important bug in the implementation of step 2.
• The criterion of Tibshirani et al. based on the GAP statistic is the new default for choosing the number of clusters.
• Allows more control over the parameters of kmeans.
• The outputs of summary, print and plot are more detailed.