Either use single thread of execution or a memmap. https://github.com/ogrisel/parallel_ml_tutorial/blob/master/rendered_notebooks/06%20-%20Distributed%20Model%20Selection%20and%20Assessment.ipynb
