The summary of the reason is that this was a summer project and parallelizing the random forest algorithm at all was a big enough project.
Writing a single pass on-line algorithm was considered a bit much for the project size. Figuring out how to make multiple passes through an input split was similarly out of scope. If you have a good alternative, this would be of substantial interest because it could improve the currently limited scalability of the decision forest code. On Thu, Jul 7, 2011 at 8:20 AM, Xiaobo Gu <[email protected]> wrote: > Why can't a tree be built against a dataset resides on the disk as > long as we can read it ? >
