add mo_polynomial with tests

See merge request !82

Rework optimization routines

See merge request !81

If only allocated arrays must be deallocated Fortran does it on its own
and no destructor/final is needed.

opti_types: added sanity checks to pointer routines; added data_shape optional input to sim_data_add; added is_allocated to sim_var; added optional raise to sim_data_get_id

Maren Kaluza authored 2 months ago and Sebastian Müller committed 2 months ago

model_dp is designed for one dimensional data_sim.

Maren Kaluza authored 2 months ago and Sebastian Müller committed 2 months ago

dataObs was only used to derive the shape of the opti data.

opti types: renaming: opti_sim => sim_data, opti_sim_single => sim_var, opti_vars => variables, dimen => ndim

Sebastian Müller authored 3 months ago and Sebastian Müller committed 2 months ago

Add optimizee type

See merge request !83

Maren Kaluza authored 6 months ago and Sebastian Müller committed 2 months ago

New features:

- The eval interface only has two variables: config, opti_sim(:)
    - config contains the parameters. Also for example the indices. The
      indices are helpful for parallelization: the eval would only use a
      subset of domains.
    - opti_sim is an array that will have the size nDomains in case of
      mHM. Each element of the array is then of type opti_sim and can
      have an array of different optimization variables 'opti_vars'.
- opti_vars replaces specific optimization variables like runoff, et,
  tws, etc. You can add an arbitrary variable with dimension 1, 2, 3, 4
  or 5 with an arbitrary name like "runoff", "et", "tws", etc.
    - optivars contains:
      - add: to add a variable. -- This might not work correctly if
	a second variable is added and so on.
      - has: by name, if a variable is added.
      - allocate: allocates the space for the variable. It cannot be
	done through a pointer.
      - set_pointer: sets a pointer to the allocated space of a
	variable.

Changes of plan:

- The opti_sim input variable in the eval interface was not supposed to
  be an array. But it was more convenient that way and would have added
  another complicated layer to the code otherwise.
- There was not supposed to be a procedure 'allocate' in opti_sim. But
  int turned out, associated arrays cannot be allocated through
  pointers.

Maren Kaluza authored 11 months ago and Sebastian Müller committed 2 months ago

instead of many optization variables the eval_interfaces now gets a
collection of all optimization variables in one. By this, the
eval_interface is short and does not need to be changed each time we
would like to optimize something else.