joezuntz

Some guidelines for statisticians/mathematicians writing samplers and minimizers if they want them to be useful to scientists.

We don't need bespoke, highly-trained algorithms: we need general methods that need minimal human supervision.
If your algorithm isn't parallel then it is unlikely to be useful (we have a lot of CPUs).
We don't have the derivative of the likelihood in most cases. It's hard enough calculating the likelihood itself.
Dumb down the paper a little for us; we spend most of our time doing other things.
If you supply some code along with your algorithm we are vastly more likely to at least give it a try.
Write your code in Python or C (or maybe C++ if you must). We need to integrate your it into more general frameworks, not statistical toolboxes.
Write your code as a library that we can call from our own programmes, not as a top-level program on its own.
Don't do I/O yourself in your code. We'll handle that, just return some samples/values/metadata/whatever.

	#Suppose you have a collection of tasks, which in this example I'll assume is just running a function f.
	#If these tasks are completely separate and independent the most then you can parallelize them easily.
	#In this gist I'll show the simplest possible way to do this using mpi4py.
	#There are better ways to do this, in particular if the tasks vary significantly in time taken to run.

	import mpi4py.MPI

	def f(i):
	"A fake task - in this case let just open a file and write a number to it"
	#open file with name based on task number

	import pylab as pl
	import numpy as np
	Z=-1jnp.array([np.exp(1.j2**i) for i in xrange(100)])
	pl.plot(0.5*Z.real, Z.imag, '-', color='#FF69B4')
	pl.show()

	import glob
	import re
	import pyparsing as pp
	import collections


	def parse_projects(text):
	#the list which will contain the located projects
	projects = []