cv

scores = cross_val_score(model,train_x,train_y,cv=5,scoring='neg_mean_squared_error')

cv或者grid_search的惯例是，会令scoring尽可能大，因为一般score是准确率这种越大越好的，而不是mse这种越小越好的。

所以mse=-neg_mean_squared_error
rmse =(-neg_mean_squared_error)**0.5