Source Code for Module ML.FeatureSelect.CMIM

 1  # $Id: CMIM.py 2 2006-05-06 22:54:39Z glandrum $ 
 2  # 
 3  #  Copyright (C) 2005  Rational Discovery LLC 
 4  #    All Rights Reserved 
 5  # 
 6  """ for the moment this is using Francois Fleuret's cmim library 
 7   to do the feature selection 
 8   
 9   Reference: F. Fleuret "Fast Binary Feature Selection with Conditional 
10              Mutual Information", J. Machine Learn. Res. 5, 1531-1535 
11              (2004) 
12     
13  """ 
14  import RDConfig 
15  import DataStructs 
16  import tempfile 
17  import os 
18  import rdFeatSelect 
19   
20 -def SelectFeatures(examples,nFeatsToPick,bvCol=1): 
21    res = rdFeatSelect.selectCMIM(examples,nFeatsToPick) 
22    if -1 in res: 
23      res = list(res) 
24      res = tuple(res[:res.index(-1)]) 
25    return res 
26   
27 -def _SelectFeatures(examples,nFeatsToPick,bvCol=1): 
28    nPts = len(examples) 
29    nFeats = examples[0][bvCol].GetNumBits() 
30   
31    exe = os.path.join(RDConfig.RDBaseDir,'External','cmim-1.0','cmim.exe') 
32    if not os.path.exists(exe): 
33      raise ValueError,'could not find cmim executable %s'%exe 
34     
35    inFname = tempfile.mktemp('.dat') 
36    outFname = inFname + '.out' 
37    inF = open(inFname,'w+') 
38    print >>inF,nPts,nFeats 
39    for row in examples: 
40      print >>inF,row[bvCol].ToBitString() 
41      print >>inF,row[-1] 
42    inF.close() 
43    inF = None 
44   
45    os.spawnlp(os.P_WAIT,exe,exe,'--nb-features',str(nFeatsToPick),'--train', 
46              inFname,outFname) 
47   
48    inD = open(outFname,'r') 
49    inL = inD.readline() 
50    nCreated = int(inL) 
51    inL = inD.readline() 
52    res = [] 
53    splitL = inL.split(' ') 
54    for i in range(nFeatsToPick): 
55      res.append(int(splitL[i])) 
56    inD.close() 
57    inD = None 
58     
59    os.unlink(inFname) 
60    os.unlink(outFname) 
61   
62    return res 
63