As of 0.98.9, only "zero-imputation" is implemented (as the default). This is sensible for univariate statistics but not multivariate statistics.

It would be helpful to have at least two other imputation methods:

• "Centering-imputation" for multivariate statistics:
  • It is sensible to use the median intensity for the feature since:
    • It makes no assumption about the shape of the distribution
    • It substitutes either one observed value for an odd number of observations.
    • It substitutes the average of two observed values for an even number of observations.
• "No imputation" to defer the decision regarding imputation until it is known whether the data will be used for univariate or multivariate statistics.