Newmu

import os
import numpy as np
from matplotlib import pyplot as plt
from time import time

from foxhound import activations
from foxhound import updates
from foxhound import inits
from foxhound.theano_utils import floatX, sharedX

mangecoeur

Easy parallel python with concurrent.futures

As of version 3.3, python includes the very promising concurrent.futures module, with elegant context managers for running tasks concurrently. Thanks to the simple and consistent interface you can use both threads and processes with minimal effort.

For most CPU bound tasks - anything that is heavy number crunching - you want your program to use all the CPUs in your PC. The simplest way to get a CPU bound task to run in parallel is to use the ProcessPoolExecutor, which will create enough sub-processes to keep all your CPUs busy.

We use the context manager thusly:

with concurrent.futures.ProcessPoolExecutor() as executor:

Slater-Victoroff

import numpy as np

def transfer_function(x, y):
    return np.power(np.prod(x, axis=1)[:, None] * np.prod(y, axis=0), 1./x.shape[1])

def gnn(c):
    return  normalize([np.random.random(c[i] * c[i + 1]).reshape((c[i], c[i + 1])) for i in range(len(c) - 1)])

def predict(weights, input_vector, reverse=False):
    current_net = [input_vector] + weights

Slater-Victoroff

from collections import Counter
import cPickle as pickle
import random
import itertools
import string

def words(entry):
    return [word.lower().decode('ascii', 'ignore') for word in entry.split()]

def letters(entry):

mblondel

"""
(C) August 2013, Mathieu Blondel
# License: BSD 3 clause

This is a Numba-based reimplementation of the block coordinate descent solver
(without line search) described in the paper:

    Block Coordinate Descent Algorithms for Large-scale Sparse Multiclass
    Classification.  Mathieu Blondel, Kazuhiro Seki, and Kuniaki Uehara.
    Machine Learning, May 2013.