nn_cross.py

   1 #!/usr/bin/python
   2 import sys
   3 import os
   4 import subprocess
   5 from gostyle import *
   6 from math import sqrt
   7 import numpy
   8
   9 from data_about_players import Data
  10
  11 class NeuralNet:
  12         def __init__( self, filename ):
  13                 s = "./gnet/gnet_train -l 3 -n 30 -p 1000 -e 0.0005 -o gonet.net ./"+filename
  14                 args = s.split()
  15                 ret = subprocess.call(args)
  16                 print ret
  17                 s = "./gnet/gnet_run gonet.net"
  18                 args = s.split()
  19                 self.p = subprocess.Popen(args, stdin=subprocess.PIPE, stdout=subprocess.PIPE)
  20         def __call__(self, vector):
  21                 self.p.stdin.write(' '.join([str(a) for a in vector]) + '\n')
  22                 a = self.p.stdout.readline()
  23                 return [ float(num) for num in a.split()]
  24
  25 if __name__ == '__main__':
  26         main_pat_filename = Data.main_pat_filename
  27         num_features = 400
  28         #k = 5
  29         player_vector = Data.questionare_total
  30         players_ignore = [ "Yi Ch'ang-ho 2004-" ]#, "Fujisawa Hideyuki","Yuki Satoshi", "Otake Hideo", "Yi Ch'ang-ho 2005+","Takao Shinji","Hane Naoki","Kobayashi Koichi" ]
  31         players_all = [ p for p in player_vector.keys() if p not in players_ignore ]
  32
  33         ### Object creating input vector when called
  34         print "Creating input vector generator from main pat file:", main_pat_filename
  35         print
  36         i = InputVectorGenerator(main_pat_filename, num_features)
  37
  38         # Create list of input vectors
  39         input_vectors = []
  40         for name in players_all:
  41                 input_vectors += [i(Data.pat_files_folder + name)]
  42
  43         #print '"%s"'%(players_all[2],)
  44         #print input_vectors[2]
  45
  46         if len(input_vectors) == 0:
  47                 print >>sys.stderr, "No reference vectors."
  48                 sys.exit()
  49
  50         ### PCA example usage
  51         # Change this to False, if you do not want to use PCA
  52         use_pca = True
  53         if use_pca:
  54                 # Create PCA object, trained on input_vectors
  55                 print >>sys.stderr, "Running PCA."
  56                 pca = PCA(input_vectors, reduce=True)
  57                 # Perform a PCA on input vectors
  58                 input_vectors = pca.process_list_of_vectors(input_vectors)
  59                 # Creates a Composed object that first generates an input vector
  60                 # and then performs a PCA analysis on it.
  61                 i = Compose(i, pca)
  62
  63         ### n/4-fold cross validation
  64         #bounds = random.sample(range(1,len(players_all)), len(players_all) / 10 )
  65         bounds=[]
  66         for x in range(1,len(players_all)/4):
  67                 bounds += [4*x for _ in [1] if 4*x < len(players_all)]
  68         if not bounds:
  69                 print >>sys.stderr, "Pop too small."
  70                 sys.exit()
  71         bounds.sort()
  72
  73         r = Rescale(-1.0,1.0)
  74         errs=[]
  75         es=[]
  76         sentinel=len(players_all)
  77         prev=0
  78         for b in bounds+[sentinel]:
  79                 validation_set = range(prev, b)
  80                 reference_set = range(0,prev) + range(b,sentinel)
  81                 print "Reference set :",
  82                 for pr in range(0, prev):
  83                         print "R",
  84                 for pr in validation_set:
  85                         print "_",
  86                 for pr in range(b, sentinel):
  87                         print "R",
  88                 print
  89                 prev = b
  90
  91                 ### Object creating output vector when called;
  92                 data =[]
  93                 for index in reference_set:
  94                         data.append( (input_vectors[index], r(player_vector[players_all[index]])) )
  95
  96
  97                 ### We can enlarge the data set by adding linear combinations of input and output vectors
  98                 use_lin_combinations = True
  99                 if use_lin_combinations:
 100                         data += Combinator().combine(data)
 101
 102                 print_set_to_file(data,'nn_cross.data')
 103
 104                 nn = NeuralNet('nn_cross.data')
 105                 # Create list of output vectors using weighted kNN algorithm approximating output_vector
 106                 def rand_vect(k):
 107                         return list(10*numpy.random.random(k))
 108                 output_vectors = [ nn(input_vectors[index]) for index in validation_set ]
 109                 #output_vectors = [ r(rand_vect(4)) for index in validation_set ]
 110                 desired_vectors = [ r(player_vector[players_all[index]]) for index in validation_set ]
 111
 112                 for vec_set,text in [(output_vectors, "Output: "), (desired_vectors, "Desired:")]:
 113                         print text,
 114                         for o in vec_set:
 115                                 for x in o:
 116                                         print "%02.3f"%(x,),
 117                                 print "; ",
 118                         print
 119
 120                 for o,d in zip(output_vectors, desired_vectors):
 121                         err = 0.0
 122                         for x,y in zip(o,d):
 123                                 e = (1.0*x-1.0*y)**2
 124                                 es += [e]
 125                                 err += e
 126                         errs += [err]
 127
 128                 #for e in errs[-4:]:
 129                 #       print "%2.3f"%(e,),
 130                 print
 131
 132
 133         print "Total square err: %2.3f"%( sum(errs),)
 134         mean = numpy.array(errs).mean()
 135         print "Mean square err per player: " + u"%2.3f ( = sd \u00B1 %2.3f)  "%(mean, sqrt(mean))
 136         mean = numpy.array(es).mean()
 137         print "Mean square err per style:  " + u"%2.3f ( = sd \u00B1 %2.3f)  "%(mean, sqrt(mean))
 138         print
 139         print "Players sorted by mean square error:"
 140         p = zip(errs, players_all)
 141         p.sort()
 142         for err, name in p:
 143                 print "%2.3f %s"%(err,name)
 144                 #print "%s"%(name,)
 145         sys.exit()