Consider enabling the use of arrays (and matrix multiplications) in the definition of dependant parameters

[rgavazzi]

Assuming you know beforehand that magnitude covariance (between n bands, say) are well captured by multivariate normal distribution N(v0, C) of central values v0, and covariance matrix C, and A^-1 is a lower triangular Choleski factorisation of C, it would be nice to enable the following definition of priors (this currently fails!).

# v0 and A, previously defined vector, and matrix, respectively

ZPT=23.9
flux={}
mag={}
centered_mag = {}
for band,group in mesgroup:
    flux[i] = get_flux_parameter()
    mag[i] = DependentParameter(lambda f: -2.5 * np.log10(f) + ZPT, flux[i] )
    centered_mag[i] = DependentParameter(lambda m:  m - v0[i], mag[i] )

eigen_mag = {}
for i in range(len(mag)):
  eigen_mag[i] = DependentParameter( lambda x: np.matmul(A,x)[i], centered_mag )
  add_prior( eigen_mag[i], 0.0, 1.0 )
  add_output_column('KK_'+str(i), eigen_mag[i])

Ideally, one would like to enable a definition of correlated multivariate Normal priors as simply as the following:

add_prior( mag, v0, C )