Minimizing Copstress to obtain a clustered MDS configuration with given hyperparameters theta.
copstressMin(
delta,
kappa = 1,
lambda = 1,
nu = 1,
theta = c(kappa, lambda, nu),
type = c("ratio", "interval", "ordinal"),
ties = "primary",
weightmat = 1 - diag(nrow(delta)),
ndim = 2,
init = NULL,
stressweight = 0.975,
cordweight = 0.025,
q = 1,
minpts = ndim + 1,
epsilon = 10,
dmax = NULL,
rang,
optimmethod = c("NelderMead", "Newuoa", "BFGS", "SANN", "hjk", "solnl", "solnp",
"subplex", "snomadr", "hjk-Newuoa", "hjk-BFGS", "BFGS-hjk", "Newuoa-hjk", "cmaes",
"direct", "direct-Newuoa", "direct-BFGS", "genoud", "gensa"),
verbose = 0,
scale = c("sd", "rmsq", "std", "proc", "none"),
normed = TRUE,
accuracy = 1e-07,
itmax = 5000,
stresstype = c("stress-1", "stress"),
...
)A list with the components
delta: the original transformed dissimilarities
obsdiss: the explicitly normed transformed dissimilarities (which are approximated by the fit)
confdist: the fitted distances
conf: the configuration to which the scaling of argument scale was applied
confo: the unscaled but explicitly normed configuration returned from the fitting procedure. Scaling applied to confo gives conf.
par, pars : the theta vector of powers tranformations (kappa,lambda,nu)
niter: number of iterations of the optimizer.
stress: the square root of explicitly normalized stress (calculated for confo).
spp: stress per point
ndim: number of dimensions
model: Fitted model name with optimizer
call: the call
nobj: the number of objects
type, loss, losstype: stresstype
stress.m: The stress used for copstress. If stresstype="stress-1" this is like $stress else it is stress^2
stress.en: another ways to calculate the stress
deltaorig: the original untransformed dissimilarities
copstress: the copstress loss value
resmat: the matrix of residuals
weightmat: the matrix of untransformed weights
OC: the (normed) OPTICS Cordillera object (calculated for scaled conf)
OCv: the (normed) OPTICS Cordillera value alone (calculated for scaled conf)
optim: the object returned from the optimization procedure
stressweight, cordweight: the weights of the stress and OC respectively (v_1 and v_2)
optimmethod: The solver used
type: the type of MDS fitted
numeric matrix or dist object of a matrix of proximities
power transformation for fitted distances
power transformation for proximities
power transformation for weights
the theta vector of powers; the first is kappa (for the fitted distances if it exists), the second lambda (for the observed proximities if it exist), the third is nu (for the weights if it exists) . If less than three elements are is given as argument, it will be recycled. Defaults to 1 1 1. Will override any kappa, lmabda, nu parameters if they are given and do not match
what type of MDS to fit. Currently one of "ratio", "interval" or "ordinal". Default is "ratio".
the handling of ties for ordinal (nonmetric) MDS. Possible are "primary" (default), "secondary" or "tertiary".
(optional) a matrix of nonnegative weights; defaults to 1 for all off diagonals
number of dimensions of the target space
(optional) initial configuration
weight to be used for the fit measure; defaults to 0.975
weight to be used for the cordillera; defaults to 0.025
the norm of the cordillera; defaults to 1
the minimum points to make up a cluster in OPTICS; defaults to ndim+1
the epsilon parameter of OPTICS, the neighbourhood that is checked; defaults to 10
The winsorization limit of reachability distances in the OPTICS Cordillera. If supplied, it should be either a numeric value that matches max(rang) or NULL; if NULL it is found as 1.5 times (for kappa >1) or 1 times (for kappa <=1) the maximum reachbility value of the power torgerson model with the same lambda. If dmax and rang are supplied and dmax is not max(rang), a warning is given and rang takes precedence.
range of the reachabilities to be considered. If missing it is found from the initial configuration by taking 0 as the lower boundary and dmax (see above) as upper boundary. See also cordillera
What optimizer to use? Choose one string of 'Newuoa' (from package minqa), 'NelderMead', 'hjk' (Hooke-Jeeves algorithm from dfoptim), 'solnl' (from nlcOptim), 'solnp' (from Rsolnp), 'subplex' (from subplex), 'SANN' (simulated annealing), 'BFGS', 'snomadr' (from crs), 'genoud' (from rgenoud), 'gensa' (from GenSA), 'cmaes' (from cmaes) and 'direct' (from nloptr). See the according R packages for details on these solvers. There are also combinations that proved to work well good, like 'hjk-Newuoa', 'hjk-BFGS', 'BFGS-hjk', 'Newuoa-hjk', 'direct-Newuoa' and 'direct-BFGS' . Usually hjk, BFGS, newuoa, subplex and solnl work rather well in an acceptable time frame (depending on the smoothness of copstress). Default is 'hjk-Newuoa'.
numeric value hat prints information on the fitting process; >2 is very verbose
Allows to scale the configuration for the OC (the scaled configuration is also returned as $conf). One of "none" (so no scaling), "sd" (configuration divided by the highest standard deviation of the columns), "std" (standardize all columns !NOTE: This does not preserve the relative distances of the optimal config), "proc" (procrustes adjustment to the initial fit) and "rmsq" (configuration divided by the maximum root mean square of the columns). Default is "sd".
should the cordillera be normed; defaults to TRUE
numerical accuracy, defaults to 1e-7
maximum number of iterations. Defaults to 5000. If itmax is (too) small, some optimizers will print warnings. For example, for optimizers using NEWUOA, an iteration number of 10*length(par)^2 is recommended. The number of parameters to optimize over for the COPS problem is number of objects * target space dimensions and can grow large very quickly, so being able to live with these warnings is probably a good idea.
which stress to use in the copstress. Defaults to stress-1. If anything else is set, explicitly normed stress which is (stress-1)^2. Using stress-1 puts more weight on MDS fit.
additional arguments to be passed to the optimization procedure
dis<-as.matrix(smacof::kinshipdelta)
#Copstress with equal weight to stress and cordillera
res1<-copstressMin(dis,stressweight=0.5,cordweight=0.5,
itmax=1000) #use higher itmax about 10000
res1
summary(res1)
plot(res1) #super clustered
Run the code above in your browser using DataLab