prefetch-generator 1.0.3

Based on http://stackoverflow.com/questions/7323664/python-generator-pre-fetch

This is a single-function package that makes it possible to transform any generator into a BackgroundGenerator which computes any number of elements from the generator ahead, in a background thread.

It is quite lightweight, but not entirely weightless.

The BackgroundGenerator is most useful when you have a GIL releasing task which might take a long time to complete (e.g. Disk I/O, Web Requests, pure C functions, GPU processing, …), and another task which takes a similar amount of time, but is dependent on the results of the first task (e.g. Computationally intensive processing of data loaded from disk).

Normally these two tasks will constantly wait for one another to finish. If you make one of these tasks a BackgroundGenerator (see examples below), they will work in parallel, potentially saving up to 50% of execution time (definitely less in practice).

We personally use the BackgroundGenerator when iterating over minibatches of data for deep learning with tensorflow and theano ( lasagne, blocks, raw, etc.).

Quick usage example (ipython notebook) - https://github.com/justheuristic/prefetch_generator/blob/master/example.ipynb

This package contains two objects:

• The Class BackgroundGenerator(generator [,max_prefetch=1])

• The decorator @prefetch([max_prefetch=1])

the usage is either

#for item in BackgroundGenerator(my_generator): # do_stuff(item)

or

#@prefetch() #def my_generator(some_param): # while True: # X = read_heavy_file() # y = wget_from_cornhub() # do_pretty_much_anything(some_param) # yield X, y

More details are written in the BackgroundGenerator doc: See help(BackgroundGenerator)