Add parameters for hyperparameterisation

[rcannood]

I propose adding this section:

config.vsh.yaml:

argument_groups:
  - name: Parameters
    arguments:
      - type: integer
        name: --foo
        info:
          optimize:
            type: linear
            lower: 10
            upper: 20
        default: 10

And then in the code:

out = harmonypy.run_harmony(
    data_mat=adata_to_correct.layers["preprocessed"],
    meta_data=adata_to_correct.obs,
    vars_use="batch_str",
    foo=par["foo"],
)

Sometimes you need an exponential uniform distribution

argument_groups:
  - name: Parameters
    arguments:
      - type: double
        name: --foo
        info:
          optimize:
            type: expuniform
            lower: 0.001
            upper: 0.1
        default: 0.01

[ghar1821]

can we also add a step value so it increases the value at this "step" value? i.e. if lower = 10, upper = 20, having step = 5 will increase the value by 5 each time rather than 1, resulting in 10, 15, 20, instead of 10,11,12,...,20.

[ghar1821]

Putting in questions based on discussion with @KatrienQ .

is it possible to specify values for parameters that take string as value? e.g., cyCombine's norm_method can accept either scale or rank.

can we also add a step value so it increases the value at this "step" value? i.e. if lower = 10, upper = 20, having step = 5 will increase the value by 5 each time rather than 1, resulting in 10, 15, 20, instead of 10,11,12,...,20.

Also, if it is possible to set steps, we can then set extreme values for some parameters. E.g., setting cytonorm's metacluster to 1 and n_quantiles to 2.