knn.py

   1 #!/usr/bin/python
   2 import sys
   3 from gostyle import *
   4 from math import sqrt
   5
   6 from data_about_players import Data
   7
   8 class KNNOutputVectorGenerator(VectorGenerator):
   9         """ k-NearestNeighbour output vector generator."""
  10         def __init__(self, ref_dict, k=5, weight_param=0.8, dist_mult=10):
  11                 """
  12                         ref_dict is a dictionary of refence input/output vectors.
  13                         e.g. ref_dict= { (1.0,2.0):(9.0,16.0,21.0)
  14                 """
  15                 self.ref_dict = ref_dict
  16                 self.k = k
  17                 self.weigth_param = weight_param
  18                 self.dist_mult = dist_mult
  19         def __call__(self, player_vector):
  20                 distance=[]
  21                 for ref_vec in self.ref_dict.keys():
  22                         distance.append((self.distance(ref_vec, player_vector), ref_vec))
  23                 distance.sort()
  24
  25                 #for p,v in distance:
  26                 #       print "%2.3f"%(float(p),),
  27                 #print
  28                 ref_output_vecs = [ self.ref_dict[b] for a,b in distance[:self.k] ]
  29                 coefs = [ self.weight_fc(a) for a,b in distance[:self.k] ]
  30
  31                 return linear_combination(ref_output_vecs, coefs)
  32         def weight_fc(self, distance):
  33                 return self.weigth_param ** (distance)
  34         def distance(self, vec1, vec2):
  35                 if len(vec1) != len(vec2):
  36                         raise RuntimeError("Dimensions of vectors mismatch.")
  37                 ### the 10* multiplicative constant is empirically determined for correct scaling
  38                 return self.dist_mult * sqrt(sum([ (float(a) - float(b))**2 for a,b in zip(vec1,vec2)]))
  39
  40
  41 if __name__ == '__main__':
  42         main_pat_filename = Data.main_pat_filename
  43         filename_play_other = 'knn_other.data'
  44         filename_play_ref = 'knn_ref.data'
  45         filename_play_ref_orig = 'knn_ref_orig.data'
  46         num_features = 300
  47         k = 4
  48         player_vector = Data.questionare_total
  49         players_ignore = [ "Yi Ch'ang-ho 2004-", "Yi Ch'ang-ho"] #,"Takao Shinji","Hane Naoki","Kobayashi Koichi" ]
  50
  51         players_all = [ p for p in Data.players_all if p not in players_ignore ]
  52         players_ref = [ p for p in player_vector if p not in players_ignore ]
  53         players_other =  [ x for x in players_all if x not in players_ref ]
  54
  55         ### Object creating input vector when called
  56         print >>sys.stderr, "Creating input vector generator from main pat file:", main_pat_filename
  57         i = InputVectorGenerator(main_pat_filename, num_features)
  58
  59         # Create list of input vectors
  60         input_vectors_ref = []
  61         for name in players_ref:
  62                 input_vectors_ref += [i(Data.pat_files_folder + name)]
  63         input_vectors_other = []
  64         for name in players_other:
  65                 input_vectors_other += [i(Data.pat_files_folder + name)]
  66
  67         if len(input_vectors_ref) == 0:
  68                 print >>sys.stderr, "No reference vectors."
  69                 sys.exit()
  70         if len(input_vectors_other) == 0:
  71                 print >>sys.stderr, "No vectors to process."
  72                 sys.exit()
  73
  74         ### PCA example usage
  75         # Change this to False, if you do not want to use PCA
  76         use_pca = False
  77         if use_pca:
  78                 # Create PCA object, trained on input_vectors
  79                 print >>sys.stderr, "Running PCA."
  80                 pca = PCA(input_vectors_ref + input_vectors_other, reduce=True)
  81                 # Perform a PCA on input vectors
  82                 input_vectors_ref = pca.process_list_of_vectors(input_vectors_ref)
  83                 input_vectors_other = pca.process_list_of_vectors(input_vectors_other)
  84                 # Creates a Composed object that first generates an input vector
  85                 # and then performs a PCA analysis on it.
  86                 i = Compose(i, pca)
  87
  88         ### Object creating output vector when called;
  89         ref_dict = {}
  90         for name, input_vector in zip(players_ref, input_vectors_ref):
  91                 ref_dict[tuple(input_vector)] = player_vector[name]
  92
  93         oknn = KNNOutputVectorGenerator(ref_dict, k=k)
  94
  95         # Create list of output vectors using weighted kNN algorithm approximating output_vector
  96         output_vectors_other = [ oknn(input_vector) for input_vector in input_vectors_other ]
  97         output_vectors_ref = [ oknn(input_vector) for input_vector in input_vectors_ref ]
  98
  99         def print_me(names, vecs, where):
 100                 if len(names) != len(vecs):
 101                         raise RuntimeError("Dimensions of vectors mismatch.")
 102
 103                 print >>sys.stderr, "Saving output_vectors to file:", where
 104                 f = open(where, 'w')
 105                 for i in xrange(len(names)):
 106                         name_to_print = '_'.join(names[i].split())
 107                         print_vector([name_to_print] + list(vecs[i]), f)
 108                 f.close()
 109
 110         print_me(players_ref, [player_vector[name] for name in players_ref], filename_play_ref_orig)
 111         print_me(players_ref, output_vectors_ref, filename_play_ref)
 112         print_me(players_other, output_vectors_other, filename_play_other)
 113
 114         print >> sys.stderr, "\nNow plot that in Gnuplot by:"
 115         #print >> sys.stderr, 'set xrange[0:%d] ; set yrange[0:%d]'%(size,size)
 116         print >> sys.stderr, 'set xtics 1 ; set ytics 1'
 117         print >> sys.stderr, 'set grid ; set size square'
 118         print >> sys.stderr, 'plot "%s" using 2:3:1 with labels font "arial,11" point lt 10 pt 4 left, "%s" using 2:3:1 with labels font "arial,11" point lt 12 pt 4 left'%(filename_play_other, filename_play_ref)
 119