tdmPreSFA.train uses package rSFA-package. It is assumed that classification for the variable
contained in column response.var is done. SFA seeks features in an expanded function space for which
the intra-class variation w.r.t. response.var is as low as possible.
tdmPreSFA.train(dset, response.var, opts)the data frame with training (and test) data.
the response variable for classification.
a list from which we need here the following entries:
PRE.SFA: [ "linear" | "2nd" | "none" ] which stands for [ 1st | 2nd degree monomial SFA | no SFA ]
PRE.SFA.REPLACE: [T] =T: replace the original numerical columns with the SFA columns; =F: add the SFA columns
PRE.SFA.npc: if >0, then add for the first PRE.SFA.npc PCs the monomials of degree 2 (see tdmPreAddMonomials)
PRE.SFA.PPRANGE: [11] number of inputs after preprocessing, they enter into expansion
PRE.SFA.ODIM: [5] number of SFA output dimensions (slowest signals) to return
PRE.SFA.numericV vector with all column names in dset which are input for SFA. These columns may contain *numeric* values only.
sfa, a list with entries:
the input data frame dset with columns numeric.variables replaced or extended (depending on opts$PRE.SFA.REPLACE)
by the SFA components with names SF1, SF2, ...
and with optional monomial columns added, if PRE.SFA.npc>0
the new numeric column names of dset, i.e. SFA components, monomials (and optionally
PRE.SFA.numericV, if opts$PRE.SFA.REPLACE==F)
a list with the items opts (sfaOpts), matrices DSF and SF and many others, as returned from
sfaStep