so samples_subspace is now a unit hypercube samping np-r dimensions
Now define parameter correlation matrix relating the hypercube to the dimensionality of the model:
e.g. if there are 5 parameters, parameters 1 and 2 are correlated, and parameters 4 and 5 are correlated:
cp=np.array([[1,1,0,0,0],[0,0,1,0,0],[0,0,0,1,1]])
constructing the full unit matrix as the dot product.
samples_unit=np.dot(samples_subspace,cp)
Need to think a bit more about how to generalise this...
Allow multiple parameters to be perfectly correlated in latin hypercube sampling.
possible approach - allow 'r' dimensions out of a total 'np' to be jointly sampled
in get_samples_from_distro_latin(self, numvalues), allow
so samples_subspace is now a unit hypercube samping np-r dimensions
Now define parameter correlation matrix relating the hypercube to the dimensionality of the model:
e.g. if there are 5 parameters, parameters 1 and 2 are correlated, and parameters 4 and 5 are correlated:
cp=np.array([[1,1,0,0,0],[0,0,1,0,0],[0,0,0,1,1]])
constructing the full unit matrix as the dot product.
samples_unit=np.dot(samples_subspace,cp)
Need to think a bit more about how to generalise this...