pca.py

   1 #!/usr/bin/python
   2 """
   3 This code creates input vectors and performs PCA on it. Each pca'd vector is then printed along with
   4 the player name, suitable e.g. to plot using gnuplot.
   5 """
   6 # not true currently
   7 """
   8 OUTPUT FORMAT
   9 player_name first_principal_component_of_player's_input_vector second_principal_component ...
  10 second_player_name ...
  11 ...
  12 """
  13 import sys
  14 from gostyle import print_vector, OccurenceVectorGenerator, Rescale, PlayerStrategyIdentificator, PCA, InputVectorGenerator
  15 from itertools import izip, count
  16 from data_about_players import Data
  17
  18 if __name__ == '__main__':
  19         main_pat_filename = Data.main_pat_filename
  20         num_features = 500
  21         filename_pca = 'pca.data'
  22         players_ignore = [ 'Cho Tae-hyeon', 'Shao Zhenzhong', 'Wu Songsheng', 'Honinbo Shusaku',  'Kuwahara Shusaku', 'Yasuda Shusaku', 'Go Seigen', 'Suzuki Goro', 'Jie Li' ] #, 'Cho Chikun', 'Takemiya Masaki']
  23         players_all = Data.players_all
  24         players = [ p for p in players_all if p not in players_ignore ]
  25         #players = Data.player_vector.keys()
  26
  27         ### Objects creating input and output vectors when called
  28         print >>sys.stderr, "Creating input vector generator from main pat file:", main_pat_filename
  29         ivg = InputVectorGenerator(main_pat_filename, num_features)
  30
  31         # Create pairs of (input vector, player name)
  32         input_vectors = []
  33         for name in players:
  34                 #input_vectors += [ivg( Data.pat_files_folder + name)]
  35                 input_vectors += [[float(occ) for occ in ivg(Data.pat_files_folder + name)]]
  36
  37         if len(input_vectors) == 0:
  38                 print >>sys.stderr, "No input vectors.", main_pat_filename
  39                 sys.exit()
  40
  41         # Create PCA object, trained on input_vectors
  42         pca = PCA(input_vectors, output_dim=10)
  43         #pca = PCA(input_vectors, reduce=True)
  44
  45         # Perform a PCA on input vectors
  46         input_vectors = pca.process_list_of_vectors(input_vectors)
  47
  48         ### Now we rescale vectors, so that each component fits on -1.0 to 1.0
  49         ### this makes a very nice plot!!
  50         r = Rescale(-1.0,1.0)
  51
  52         ### Normalize each component separately
  53         # We need to transpose input_vectors - a list of per-player-vector-of-pca-component
  54         # to get list of vectors of per-component-vector-of-player-data
  55         def transpose(list_of_vectors):
  56                 return zip(*list_of_vectors)
  57         input_vectors = transpose([ r(vector) for vector in transpose(input_vectors)])
  58
  59
  60         print >> sys.stderr, "Writing output to file: ", filename_pca
  61         # prints vectors along with player names
  62         f=open(filename_pca, 'w')
  63         for name,vector in izip(players, input_vectors):
  64                 # Substitute ' ' by '_' to allow for gnuplot plotting (recognizing columns correctly)
  65                 name_to_print = '_'.join(name.split())
  66                 for p, i in izip(vector, count()):
  67                         print >> f, name_to_print, i+1, p
  68         f.close()
  69
  70         print "\nProjection info:"
  71         P = pca.get_projection_info()
  72         for y in xrange(1, P.shape[0]):
  73                 for x in xrange(1, P.shape[1]):
  74                         print y, x, P[y,x], ivg.ovg.stringof(x)
  75         #print P
  76
  77         print >> sys.stderr, "\nNow print that by:"
  78         print >> sys.stderr, 'gnuplot> set xrange[1:%d]'%(pca.pca.output_dim+1)
  79         print >> sys.stderr, 'plot "%s" using 2:3:1 with labels font "arial,10" left point pt 4 offset 1,0'%(filename_pca,)