#!/usr/bin/python -*- coding: utf-8 -*-
#
# Merlin - Almost Native Python Machine Learning Library: ML utilities
#
# Copyright (C) 2014-2015 alvations
# URL:
# For license information, see LICENSE.md
import numpy as np
from itertools import izip
import operator
def sort_dic_by_value(dic,reverse=False):
return sorted(dic.iteritems(),
key=operator.itemgetter(1),
reverse=reverse)
def dict_max(dic):
"""
Returns maximum value of a dictionary.
"""
aux = dict(map(lambda item: (item[1],item[0]),dic.items()))
if aux.keys() == []:
return 0
max_value = max(aux.keys())
return max_value,aux[max_value]
def perp_2d(a):
"""
Gets a perpendicualar line in 2D
"""
res = 1./a
res = res[:,] * [-1,1]
return res
def l2norm(a):
"""
L2 normalize and array
"""
return np.sqrt(l2norm_squared(a))
def l2norm_squared(a):
"""
L2 normalize squared
"""
value = 0
for i in xrange(a.shape[1]):
value += np.dot(a[:,i],a[:,i])
return value
def normalize_array(a,direction="column"):
"""
Normalizes an array to sum to one, either column wise,
or row wise or the full array.
*Directions*
Column-wise - 0 default
Row-wise - 1 default
All - 2 default
"""
b = a.copy()
if(direction == "column"):
sums = np.sum(b,0)
return np.nan_to_num(b/sums)
elif(direction == "row"):
sums =np.sum(b,1)
return np.nan_to_num((b.transpose() / sums).transpose())
elif(direction == "all"):
sums = np.sum(b)
return np.nan_to_num(b / sums)
else:
print "Error non existing normalization"
return b
