EIDORS: Electrical Impedance Tomography and Diffuse Optical Tomography Reconstruction Software

Hosted by

EIDORS: Programming / Caching

Requirements

1. Caching should be 'natural'

   This part of the 'overloaded' accessor functions, so for example,

   calc_image_prior used to be
   

          image_prior= feval( inv_model.image_prior.func, inv_model);
   

   now it is (using the eidors_obj function):

          image_prior = eidors_obj('cache', inv_model, 'image_prior');
   
          if isempty(image_prior)
              image_prior= feval( inv_model.image_prior.func, inv_model);
              eidors_obj('cache', inv_model, 'image_prior', image_prior);
          end
   

   so this means that the function 'pointer' in inv_model.image_prior.func = 'np_calc_image_prior' doesn't need to know anything about possible caching.

2. Cached values should not appear when the underlying model has changed.

   This is ensured by creating an 'object repository' using the eidors_obj function. eidors objects now must be constructed using this function, either as

          demo_inv.name= 'Nick Polydorides EIT inverse';
          demo_inv.solve=       'np_inv_solve';
          demo_inv.hyperparameter= 1e-8;
          demo_inv.image_prior.func= 'np_calc_image_prior';
          demo_inv= eidors_obj('inv_model', demo_inv);
   

   or as

          demo_inv= eidors_obj( ...
               'inv_model', 'Nick Polydorides EIT inverse',...
               'solve',          'np_inv_solve', ...
               'hyperparameter', 1e-8, ...
               'func',           'np_calc_image_prior');
   

   whenever an 'object' is modified, such as

          eidors_obj('set', demo_inv, 'solve', 'NEW_SOLVER_CODE' );
   

   then all cached values are flushed.

Implementation