slow operation assumes the two x's are the same @param o

@return left index (inclusive) of interval encoding to

@param f @param regularizeTerm @param adaptions used to scan for const term (no penalty on this) @return

@return right index (exclusive) of interval encoding to

@return name of original column

@param row @return can return null (on partial vectors)

@param x model coefficients (must have length dim()) @param ei data row @param r where to write prediction results (must have length noutcomes())

from: http://martin.ankerl.com/2007/02/11/optimized-exponential-functions-for-java/ �A Fast, Compact Approximation of the Exponential Function� Nico

confirm row has complete set of non-empty independent variables @param row @return

@param base how many indices to skip (use to get per-outcome data, so base will be a multiple of encoding dimension) @param x vector soln. @param leve

@param dat data source @param serialObserver (option can be null) a cheap observer that will be applied to all data serially @param parallelObserver a

@param x model coefficients (must have length dim()) @param wantGrad @param wantHessian @param d data row @param r where to sum results @param pscratc

for test/debugging, allowed to be slow @param base how many indices to skip (use to get per-outcome data, so base will be a multiple of encoding dime

sets should have identical category key set at this point (not levels) @param o

type of adaption @return

@param outcome training outcome @param level level of variable @param pred current model predictions @param correct category @paream weight>0.0