Current 1-thread setup can raise the following problem since the sklearn models are parallelable by default: