Ансамбль деревьев

function build_forest(y, X, n_subfeatures, n_trees, partial_sampling, max_depth, min_samples_leaf, min_samples_split)
    n_samples = size(X, 1)
    trees = []
    oob_predictions = zeros(n_samples)
    oob_counts = zeros(Int, n_samples)
    
    for _ in 1:n_trees
      
        sample_idx = rand(1:n_samples, Int(floor(n_samples * partial_sampling)))
        oob_idx = setdiff(1:n_samples, sample_idx)
        
        X_sample = X[sample_idx, :]
        y_sample = y[sample_idx]
        tree = DecisionTree.build_tree(y_sample, X_sample, 
                                      n_subfeatures, max_depth, 
                                      min_samples_leaf, min_samples_split)
        
        for i in oob_idx
            if i <= n_samples
                pred = DecisionTree.apply_tree(tree, X[i, :])
                if !isnan(pred) && !isinf(pred)
                    oob_predictions[i] += pred
                    oob_counts[i] += 1
                end
            end
        end
        
        push!(trees, tree)
    end
    
    for i in 1:n_samples
        if oob_counts[i] > 0
            oob_predictions[i] /= oob_counts[i]
        else
            oob_predictions[i] = NaN
        end
    end
    
    return trees, oob_predictions
end