Milk:Machine Learning Toolkit in Python

1. # -*- coding: utf-8 -*-
   
2. # Copyright (C) 2010-2011, Luis Pedro Coelho <luis@luispedro.org>
   
3. # vim: set ts=4 sts=4 sw=4 expandtab smartindent:
   
4. #
   
5. # License: MIT. See COPYING.MIT file in the milk distribution
   
6. 
   
7. '''
   
8. Random Forest
   
9. -------------
   
10. Main elements
    
11. -------------
    
12. rf_learner : A learner object
    
13. '''
    
14. 
    
15. from __future__ import division
    
16. import numpy as np
    
17. import milk.supervised.tree
    
18. from .normalise import normaliselabels
    
19. from .base import supervised_model
    
20. from ..utils import get_nprandom
    
21. 
    
22. __all__ = [
    
23.     'rf_learner',
    
24.     ]
    
25. 
    
26. def _sample(features, labels, n, R):
    
27.     '''
    
28.     features', labels' = _sample(features, labels, n, R)
    
29.     Sample n element from (features,labels)
    
30.     Parameters
    
31.     ----------
    
32.     features : sequence
    
33.     labels : sequence
    
34.         Same size as labels
    
35.     n : integer
    
36.     R : random object
    
37.     Returns
    
38.     -------
    
39.     features' : sequence
    
40.     labels' : sequence
    
41.     '''
    
42. 
    
43.     N = len(features)
    
44.     sfeatures = []
    
45.     slabels = []
    
46.     for i in range(n):
    
47.         idx = R.randint(N)
    
48.         sfeatures.append(features[idx])
    
49.         slabels.append(labels[idx])
    
50.     return np.array(sfeatures), np.array(slabels)
    
51. 
    
52. class rf_model(supervised_model):
    
53.     def __init__(self, forest, names, return_label = True):
    
54.         self.forest = forest
    
55.         self.names = names
    
56.         self.return_label = return_label
    
57. 
    
58.     def apply(self, features):
    
59.         rf = len(self.forest)
    
60.         votes = sum(t.apply(features) for t in self.forest)
    
61.         if self.return_label:
    
62.             return (votes > (rf//2))
    
63.         return votes / rf
    
64. 
    
65. 
    
66. class rf_learner(object):
    
67.     '''
    
68.     Random Forest Learner
    
69.     learner = rf_learner(rf=101, frac=.7)
    
70.     Attributes
    
71.     ----------
    
72.     rf : integer, optional
    
73.         Nr of trees to learn (default: 101)
    
74.     frac : float, optional
    
75.         Sample fraction
    
76.     R : np.random object
    
77.         Source of randomness
    
78.     '''
    
79.     def __init__(self, rf=101, frac=.7, R=None):
    
80.         self.rf = rf
    
81.         self.frac = frac
    
82.         self.R = get_nprandom(R)
    
83. 
    
84.     def train(self, features, labels, normalisedlabels=False, names=None, return_label=True, **kwargs):
    
85.         N,M = features.shape
    
86.         m = int(self.frac*M)
    
87.         n = int(self.frac*N)
    
88.         R = get_nprandom(kwargs.get('R', self.R))
    
89.         tree = milk.supervised.tree.tree_learner(return_label=return_label)
    
90.         forest = []
    
91.         if not normalisedlabels:
    
92.             labels,names = normaliselabels(labels)
    
93.         elif names is None:
    
94.             names = (0,1)
    
95.         for i in range(self.rf):
    
96.             forest.append(
    
97.                     tree.train(*_sample(features, labels, n, R),
    
98.                                **{'normalisedlabels' : True})) # This syntax is necessary for Python 2.5
    
99.         return rf_model(forest, names, return_label)

复制代码