Local Jaccard MDS pipeline at fixed (h, k)

Run the full pipeline of local Jaccard distance, multi-dimensional scaling with sequential Mizuta modification, trajectory distance, and Ward clustering. Returns the cluster labels, the per-time configurations, the trajectory distance, the smoothed curves and the class mean curves.

Usage

ljmds.pipeline(X, t, h, k)

Arguments

X

Integer/numeric n x p matrix of 0/1 keyword occurrences. Rows are time points, columns keywords.

t

Numeric vector of length n giving the time of each row.

h

Bandwidth (standard deviation of the Gaussian kernel).

k

Number of classes.

Value

A list with class "ljmds":

• labels: integer p-vector of cluster ids,

• xs, ys: n x p matrices of MDS coordinates,

• H: p x p trajectory distance,

• f: n x p smoothed occurrence curves,

• m: n x k class mean curves,

• t, h, k, keywords.

See also

ljmds.read.csv() to load a corpus, ljmds.select() to choose (h, k) from a grid, ljmds.silhouette() for the criterion used inside ljmds.select(), plot.ljmds() for diagnostic figures, ljmds.animate() to render a GIF.