Source Code for Module nltk.metrics.spearman

 1  # Natural Language Toolkit: Spearman Rank Correlation 
 2  # 
 3  # Copyright (C) 2001-2011 NLTK Project 
 4  # Author: Joel Nothman <jnothman@student.usyd.edu.au> 
 5  # URL: <http://nltk.org> 
 6  # For license information, see LICENSE.TXT 
 7   
 8  """ 
 9  Tools for comparing ranked lists. 
10  """ 
11   
12 -def _rank_dists(ranks1, ranks2): 
13      """Finds the difference between the values in ranks1 and ranks2 for keys 
14      present in both dicts. If the arguments are not dicts, they are converted 
15      from (key, rank) sequences. 
16      """ 
17      ranks1 = dict(ranks1) 
18      ranks2 = dict(ranks2) 
19      for k, v1 in ranks1.iteritems(): 
20          try: 
21              yield k, v1 - ranks2[k] 
22          except KeyError: 
23              pass 
24   
25   
26 -def spearman_correlation(ranks1, ranks2): 
27      """Returns the Spearman correlation coefficient for two rankings, which 
28      should be dicts or sequences of (key, rank). The coefficient ranges from 
29      -1.0 (ranks are opposite) to 1.0 (ranks are identical), and is only 
30      calculated for keys in both rankings (for meaningful results, remove keys 
31      present in only one list before ranking).""" 
32      n = 0 
33      res = 0 
34      for k, d in _rank_dists(ranks1, ranks2): 
35          res += d * d 
36          n += 1 
37      try: 
38          return 1 - (6 * float(res) / (n * (n*n - 1))) 
39      except ZeroDivisionError: 
40          # Result is undefined if only one item is ranked 
41          return 0.0 
42   
43   
44 -def ranks_from_sequence(seq): 
45      """Given a sequence, yields each element with an increasing rank, suitable 
46      for use as an argument to L{spearman_correlation}. 
47      """ 
48      return ((k, i) for i, k in enumerate(seq)) 
49   
50   
51 -def ranks_from_scores(scores, rank_gap=1e-15): 
52      """Given a sequence of (key, score) tuples, yields each key with an 
53      increasing rank, tying with previous key's rank if the difference between 
54      their scores is less than rank_gap. Suitable for use as an argument to 
55      L{spearman_correlation}. 
56      """ 
57      prev_score = None 
58      rank = 0 
59      for i, (key, score) in enumerate(scores): 
60          try: 
61              if abs(score - prev_score) > rank_gap: 
62                  rank = i 
63          except TypeError: 
64              pass 
65          
66          yield key, rank 
67          prev_score = score 
68