# 1.11.4.3. Fitting additional weak-learners

from sklearn.metrics import mean_squared_error

from sklearn.datasets import make_friedman1

from sklearn.ensemble import GradientBoostingRegressor

X, y = make_friedman1(n_samples=1200, random_state=0, noise=1.0)

X_train, X_test = X[:200], X[200:]

y_train, y_test = y[:200], y[200:]

est = GradientBoostingRegressor(n_estimators=100, learning_rate=0.1,

max_depth=1, random_state=0, loss='ls').fit(X_train, y_train)

_ = est.set_params(n_estimators=200, warm_start=True) # set warm_start and new nr of trees

_ = est.fit(X_train, y_train) # fit additional 100 trees to est

print mean_squared_error(y_test, est.predict(X_test))

# 3.84...