Daniel@0: # Compute the weight vector of linear SVM based on the model file
Daniel@0: # Original Perl Author: Thorsten Joachims (thorsten@joachims.org)
Daniel@0: # Python Version: Ori Cohen (orioric@gmail.com)
Daniel@0: # Call: python svm2weights.py svm_model
Daniel@0: 
Daniel@0: import sys
Daniel@0: from operator import itemgetter
Daniel@0: 
Daniel@0: try:
Daniel@0:     import psyco
Daniel@0:     psyco.full()
Daniel@0: except ImportError:
Daniel@0:     print 'Psyco not installed, the program will just run slower'
Daniel@0: 
Daniel@0: def sortbyvalue(d,reverse=True):
Daniel@0:     ''' proposed in PEP 265, using  the itemgetter this function sorts a dictionary'''
Daniel@0:     return sorted(d.iteritems(), key=itemgetter(1), reverse=True)
Daniel@0: 
Daniel@0: def sortbykey(d,reverse=True):
Daniel@0:     ''' proposed in PEP 265, using  the itemgetter this function sorts a dictionary'''
Daniel@0:     return sorted(d.iteritems(), key=itemgetter(0), reverse=False)
Daniel@0: 
Daniel@0: def get_file():
Daniel@0:     """
Daniel@0:     Tries to extract a filename from the command line.  If none is present, it
Daniel@0:     assumes file to be svm_model (default svmLight output).  If the file 
Daniel@0:     exists, it returns it, otherwise it prints an error message and ends
Daniel@0:     execution. 
Daniel@0:     """
Daniel@0:     # Get the name of the data file and load it into 
Daniel@0:     if len(sys.argv) < 2:
Daniel@0:         # assume file to be svm_model (default svmLight output)
Daniel@0:         print "Assuming file as svm_model"
Daniel@0:         filename = 'svm_model' 
Daniel@0:         #filename = sys.stdin.readline().strip()
Daniel@0:     else:
Daniel@0:         filename = sys.argv[1]
Daniel@0: 
Daniel@0:     
Daniel@0:     try:
Daniel@0:         f = open(filename, "r")
Daniel@0:     except IOError:
Daniel@0:         print "Error: The file '%s' was not found on this system." % filename
Daniel@0:         sys.exit(0)
Daniel@0: 
Daniel@0:     return f
Daniel@0: 
Daniel@0: 
Daniel@0: 
Daniel@0: 
Daniel@0: if __name__ == "__main__":
Daniel@0:     f = get_file()
Daniel@0:     i=0
Daniel@0:     lines = f.readlines()
Daniel@0:     printOutput = True
Daniel@0:     w = {}
Daniel@0:     for line in lines:
Daniel@0:         if i>10:
Daniel@0:             features = line[:line.find('#')-1]
Daniel@0:             comments = line[line.find('#'):]
Daniel@0:             alpha = features[:features.find(' ')]
Daniel@0:             feat = features[features.find(' ')+1:]
Daniel@0:             for p in feat.split(' '): # Changed the code here. 
Daniel@0:                 a,v = p.split(':')
Daniel@0:                 if not (int(a) in w):
Daniel@0:                     w[int(a)] = 0
Daniel@0:             for p in feat.split(' '): 
Daniel@0:                 a,v = p.split(':')
Daniel@0:                 w[int(a)] +=float(alpha)*float(v)
Daniel@0:         elif i==1:
Daniel@0:             if line.find('0')==-1:
Daniel@0:                 print 'Not linear Kernel!\n'
Daniel@0:                 printOutput = False
Daniel@0:                 break
Daniel@0:         elif i==10:
Daniel@0:             if line.find('threshold b')==-1:
Daniel@0:                 print "Parsing error!\n"
Daniel@0:                 printOutput = False
Daniel@0:                 break
Daniel@0:         
Daniel@0:         i+=1    
Daniel@0:     f.close()
Daniel@0: 
Daniel@0:     #if you need to sort the features by value and not by feature ID then use this line intead:
Daniel@0:     #ws = sortbyvalue(w) 
Daniel@0:     
Daniel@0:     ws = sortbykey(w)
Daniel@0:     if printOutput == True:
Daniel@0:         for (i,j) in ws:
Daniel@0:             print i,':',j
Daniel@0:             i+=1
Daniel@0: 
Daniel@0: