Parameters tuning

  • Regarding Regularized CCA, does it exist default values for the regularization parameters?

Default values for regularization parameters are not relevant as they strongly depend on the data. These parameters have to be estimated via cross-validation (see estim.regul() function).

  • I have a ‘chol.default(Cxx) : the leading minor of order 30 is not positive definite’ error when running rcc.

This is mostly likely to occur when encountering singular matrices, where the total number of variables from both data sets is much larger than the number of samples. We suggest using regularized CCA instead (i.e. with regularization parameters lambda1 and lambda2 greater than 0) or PLS or sparse PLS in canonical mode.