get_rational function

Generate start allocation based on a priori profiles

If cluster profiles are given a priori, this function can be used to compute the conditionally optimal cluster membership matrix A which can then be used as a rational starting allocation for the (low dimensional) ADPROCLUS procedure (see adproclus and adproclus_low_dim).

get_rational(data, starting_profiles)

Arguments

• data: Object-by-variable data matrix of class matrix or data.frame.
• starting_profiles: A matrix where each row represents the profile values for a cluster. Needs to be of same dimensions as $P$.

Returns

get_rational() returns a list with the following components:

• type: A character string denoting the type of start ('Rational Start')
• A: An initial Membership matrix

Details

The function uses the same quadratic loss function and minimization method as the (low dimensional) ADPROCLUS procedure does to find the next conditionally optimal membership matrix A. (for details, see Depril et al., 2012). For the full dimensional ADPROCLUS it uses the algorithm ALS2 and not ALS1.

Warning: This function does not obtain an ADPRCOLUS model. To perform additive profile clustering, see adproclus.

Examples

# Obtain data from standard data set "Stackloss"
x <- stackloss

# Obtaining a user-defined rational start profile matrix
# (here the first 4 rows of the data)
start_allocation <- get_rational(x, x[1:4, ])$A

References

Depril, D., Van Mechelen, I., & Wilderjans, T. F. (2012). Lowdimensional additive overlapping clustering. Journal of classification, 29, 297-320.

See Also

• adproclus, adproclus_low_dim: for details about membership and profile matrices
• get_random: for generating random starts
• get_semirandom: for generating semi-random starts