Correct possibility to use a scaling factor (kappa) or not

[pierrePalud]

The package was implemented to let the possibility to the user to infer a map of scaling factors $(\kappa_n)$. For each pixel $n$, this scaling factor $\kappa_n$ scales all predicted observables $\ell$ identically, such that the forward model is $\kappa_n f_\ell(\theta_n)$.

Some parts of the current implementation force the user to use a scaling factor. For instance, the current implementations of the MyScaler and NeuralNetworkApprox classes are only valid with a scaling factor. In particular, they enforce D = D_no_kappa + 1 where D is supposed to be the total number of physical parameters, and D_no_kappa the number of physical parameters excluding the scaling factor.

Changes propositions:

• Change name from kappa to scaling_factor throughout the code for clarity, as this $\kappa$ notation is not widespread.
• Correct MyScaler class to allow one not to use a scaling factor.
• Correct NeuralNetworkApprox class to allow one not to use a scaling factor.