Actually this provides a solution for a very serious problem which is usually the main bottleneck in machine learning experimentation (in terms of speed). For me at least, this was a great read and I'm pretty sure I'll be trying this out soon.