API Change surrounding inner cv

[samihamdan]

I find the api for defining the inner cv no intuitive and would like to change it.

Current API (complete example later):

model_params = model_params=dict(
    svm__C = [1,2,3],
    cv = StratifiedKFold(3),
    search='grid',
    search_params = dict(verbose=1)
)

My Problem

As you can see we specify arguments of the search inside of search_params.
But the argument cv is set outside of seach_params in model_params.
This can and did lead to confusion for users, as cv is a param of search,
but if you set it inside of search_params it there will neither result in the correct output nor a warning or error.

Example of the Problem:

Reproducible example of current API

import seaborn as sns 
from sklearn.model_selection import StratifiedKFold

from julearn import run_cross_validation
from julearn.utils import configure_logging
configure_logging(level='INFO')

df_iris = (sns.load_dataset('iris')
           .iloc[:100,:])
X = df_iris.columns[:-1].tolist()
y = 'species'

df_iris = sns.load_dataset('iris')
df_iris = df_iris.iloc[:100,:]

model_params = model_params=dict(
    svm__C = [1,2,3],
    cv = StratifiedKFold(3),
    search='grid',
    search_params = dict(verbose=1)
)
run_cross_validation(
    X=X , y=y, data=df_iris,
    model='svm', cv=3, 
    model_params=model_params
)

can be adjusted to also show the above mentioned strange behavior by changing model_params:

model_params = model_params=dict(
    svm__C = [1,2,3],
    search='grid',
    search_params = dict(
        cv = StratifiedKFold(3), verbose=1
    )

Proposed Changes:

I would propose that we only allow setting inner cv as cv in search_params.

[fraimondo]

+1

Maybe do it as part of #145 ?