Randomized_search procedure - flow of iterations

[bgreat5]

Problem: I am trying to run random search using randomized_search provided in the Catboost.
The grid is grid = {'iterations' : [700,800,900,1000,1100,1200],
'learning_rate': np.linspace(0.01,0.5,25),
'depth': [4,5,6,7,8,9,10],
'l2_leaf_reg': [1, 3, 5, 7, 9,10] }

randomized_search_result = model1.randomized_search(grid,
X=train_dataset,
cv = 3,
partition_random_seed=10,
n_iter=2,
refit=False,
plot=True)

Catboost version: 1.0.5

CPU

Now when I run it, it first runs two iterations using random 2 samples from the grid, then using the best sample it runs 3 fold cross validation and then it trains the model. However, the sklearn's RandomsearchCV function for the same grid, performs 3 fold cross validation for each sample from the grid. Is my understanding correct or Am i missing something here?

And RandomsearchCV provides with a list of samples it has tried from the Grid, will it be possible have something similar in randomized_search?

[Sarustam]

для обучения модели установите refit=True