Is there any way to do RandomizedSearchCV from scikit-learn, when validation data does already exist as a holdout set? I have tried to concat train and validation data and define the cv parameter to split exactly where the two sets where combined, but could not find a proper syntax that is accepted by RandomizedSearchCV.
scikit-learn docu says:
cv : int, cross-validation generator or an iterable, optional
Determines the cross-validation splitting strategy.
Possible inputs for cv are:
- None, to use the default 3-fold cross validation,
- integer, to specify the number of folds in a `(Stratified)KFold`,
- An object to be used as a cross-validation generator.
- An iterable yielding train, test splits.
The last option should somehow work, I hope, but I don't know in which format I have to hand it over.
Any help is appreciated!
Suppose you have the indices of your training samples in train_indices and the indices of your test samples in test_indices. Then, it is sufficient to pass these as a tuple wrapped in a list to the cv parameter of RandomizedSearchCV. A MWE to demonstrate:
from sklearn.datasets import make_classification
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import RandomizedSearchCV
X, y = make_classification(n_samples=10)
param_distributions = {
'n_estimators': [10, 20, 30, 40]
}
train_indices = [0, 1, 2, 3, 4]
test_indices = [5, 6, 7, 8, 9]
cv = [(train_indices, test_indices)]
search = RandomizedSearchCV(
RandomForestClassifier(),
param_distributions=param_distributions,
cv=cv,
n_iter=2
)
search.fit(X, y)
This will always train and test the estimator on the same samples. If your data is stored pandas dataframes, e.g. df, use df.index.values to get the indices you need.