Add support for categorical parameter types

[xjules]

Update from meeting (13.12.2024):

One idea that was discussed is to decouple the parameter group concept and make it purely as metadata attribute (see. ScalarParameter). This would essentially make every parameter "independent". Additionally, it requires a different approach when storing parameters into storage, wherein for testing purposes we could start with either pandas or polars.

from dataclasses import dataclass


parameter_configs = List[FieldParameter, SurfaceParameter, ScalarParameter]


@dataclass
class UniformSettings:    
    name: Literal["uniform"] = uniform
    min: float  
    max: float


@dataclass
class PolarsData:
    name: Literal["polars"]
    data_set_file: Path


@dataclass 
class ScalarParameter:
    name: str
    group: None | str = None
    distribution: Union[UniformSettings, ...]
    active: bool
    input_source: Literal["design_matrix", "sampled"]
    dataset_file: Union[PolarsData, XArrayData]

@yngve-sk also suggested potentially to encode categories as integers and store the mapping into a stand-alone entry in parameters.json.

---

Currently all GEN_KW group param types assume to be numbers (float), which prevents a direct read / write of categorical data from / to the storage.

ert/src/ert/config/gen_kw_config.py

Line 318 in 05b33a5

def save_parameters(

Another issue relates to the actual design matrix group. This often contains a mixture of numerical and categorical parameters, which when being loaded from excel sheet as a DESIGN_MATRIX parameter group would then render all parameters as strings (into the storage that is). Nevertheless when storing such a parameter group, ert utilizes xr.Datasets concept of variables:

ert/src/ert/config/gen_kw_config.py

Line 325 in 05b33a5

ds = xr.Dataset(

and netcdf to actually store the parameters:

ert/src/ert/storage/local_ensemble.py

Line 789 in 05b33a5

def save_parameters(

This again is based on variables, wherein a single Variable equals a single Parameter group, which needs to be of the same type.

What we need is a support for a mixture of types in a single param group or to come up with another strategy of how to do it.