Added feature selection and gitignore

.gitignore

+# Compiled source #
+###################
+*.pyc

scripts/feature_selection/crf_wrapper.py

@@ -12,6 +12,7 @@ from optparse import OptionParser
 from threading import Thread
 
 descr = """%prog [options] corpus_dir out_dir"""
+wind = 2
 
 def go():
     parser = OptionParser(usage=descr)
@@ -86,29 +87,39 @@ def process_fold(fold, out_dir, corpus_dir, config_dir, config_name):
     os.remove(os.path.join(out_dir, "models", str(fold).zfill(2), "dict-sie.lex"))
 
 #Wybranie sąsiada
-def neightbour(v):
-    a = random.randint(0, len(v))
-    b = random.randint(0, len(v[0]))
+def neightbour(v = [], c = {}):
+    global wind
+    a = random.randint(0, len(v)+len(c))
+    b = random.randint(0, wind)
+    if a < len(v):
         v[a][b] = 1 if v[a][b] == 0 else 0
-    a = random.randint(0, len(v))
-    b = random.randint(0, len(v[0]))
+    else:
+        s_c = sorted(c)
+        s_c[a-len(v)] = 1 if s_c[a-len(v)] == 0 else 0
+        
+    a = random.randint(0, len(v)+len(c))
+    b = random.randint(0, wind)
+    if a < len(v):
         v[a][b] = 1 if v[a][b] == 0 else 0
-    return v
+    else:
+        s_c = sorted(c)
+        s_c[a-len(v)] = 1 if s_c[a-len(v)] == 0 else 0
+    return (v, c)
 
 #Obliczenie temperatury początkowej
-def tempestimation(out_dir, config_dir, config_name, corpus_dir, folds, vector):
+def tempestimation(out_dir, config_dir, config_name, corpus_dir, folds, vector, constructed):
     iterations = 100           #ilosc iteracji symulacji
     sum = 0
     results = {}                 #histogram wyników
     for i in range(iterations):
-        
+        print "Temperature estimation it:" + i
         result = f(os.path.join(out_dir, str(i).zfill(2)), config_dir, config_name, corpus_dir, folds, vector)
         sum += result
         if result not in results.keys():
             results[result] = 1
         else:
             results[result] += 1
-        vector = neightbour(vector)
+        vector, constructed = neightbour(vector, constructed)
     avg = sum / float(iterations)#obliczenie średniego wyniku
     k = 0
     deviation = 0
@@ -128,10 +139,12 @@ def get_features_number(cclfile):
     return count + 1
     
 def generate_features_txt(resultfile, vector = [], constructed = {}):
+    global wind
     out = open(resultfile, 'w+')
     feature_num = 0
-    actual_feature_num = len(vector[0]) / 2 + 1
+    actual_feature_num = wind / 2 + 1
     
+    if len(vector) > 0:
         for i in range(len(vector)):
             for j in range(len(vector[i])):
                 if vector[i][j] == 1:
@@ -139,8 +152,20 @@ def generate_features_txt(resultfile, vector = [], constructed = {}):
                     out.write("\n")
                     feature_num += 1
         out.write("\n")
-    for i in range(len(constructed)):
-        feats = constructed[i].split("%")
+    if len(constructed) > 0:
+        for key in constructed.keys():
+            if constructed[key] == 1:
+                feature_substrings = key.split("%")
+                for subs in feature_substrings:
+                    feature1 = subs.split(".")[0]
+                    feature1_num = subs.split(".")[1]
+                    if subs == feature_substrings[0]:
+                        out.write('U%02d:%%x[%s,%s]/'%(feature_num, feature1_num, feature1))
+                    elif subs == feature_substrings[-1]:
+                        out.write('%%x[%s,%s]'%(feature1_num, feature1))
+                    else:
+                        out.write('%%x[%s,%s]/'%(feature1_num, feature1))
+                feature_num += 1
                 
         out.write("\n")
     out.write("B")
@@ -168,33 +193,43 @@ def P(e, en, temp):
         return 1
 
 def main(corpus_dir, out_dir, config, window, folds):
+    global wind
+    wind = window
     config_dir = os.path.dirname(config)
     config_name = os.path.splitext(os.path.basename(config))[0]
     config_ccl = os.path.join(config_dir, config_name + ".ccl")
     
-    constructed = []
-    constructed.append("1%2")
-    generate_features_txt(os.path.join(out_dir, "config_files", config_name + "-layer1.txt"), constructed = constructed)
-    
-#    if not os.path.exists(out_dir):
-#        tools.mkdir_p(out_dir)
-#    
-#    a_vector = randomize_vector(get_features_number(config_ccl), window)
-#    temperature = tempestimation(os.path.join(out_dir, "estimation"), config_dir, config_name, corpus_dir, folds, a_vector)
-#    
-#    a_value = f(os.path.join(out_dir, "selection", "first"), config_dir, config_name, corpus_dir, folds, a_vector)
-#    b_value = 0
-#    i = 1
-#    while temperature > 0:
-#        b_vector = neightbour(a_vector)
-#        b_value = f(os.path.join(out_dir, "selection", str(i).zfill(2)), config_dir, config_name, corpus_dir, folds, b_vector)
-#        prob = P(a_value, b_value, temperature)        
-#        if b_value > a_value:
-#            a_vector = b_vector
-#        elif random.randint(0, 1) < prob:
-#            a_vector = b_vector
-#        temperatore = temperature * 0.95
-#        i += 1
+    if not os.path.exists(out_dir):
+        tools.mkdir_p(out_dir)
+    
+    vector = [[1, 1, 1, 1, 0],[1, 1, 1, 0, 0]]
+
+    constructed = {'0.-1%0.0' :0, '0.0%0.1' :0, 
+                   '1.-2%1.-1':0, '1.0%1.-1':0, '1.0%1.1':0, '1.1%1.2':0,
+                   '7.-1%8.-1':0, '7.0%8.0' :0, '7.1%8.1':0,
+                   '1.-2%1.-1%1.0' : 0, '1.-1%1.0%1.1' : 0, '1.0%1.1%1.2' : 0
+                   }
+    
+    a_vector = randomize_vector(get_features_number(config_ccl), window)
+    a_constructed = constructed
+    temperature = tempestimation(os.path.join(out_dir, "estimation"), config_dir, config_name, corpus_dir, folds, a_vector)
+    
+    a_value = f(os.path.join(out_dir, "selection", "first"), config_dir, config_name, corpus_dir, folds, a_vector)
+    b_value = 0
+    i = 1
+    while temperature > 0:
+        print "Feature selection temp:" + temperature
+        b_vector, b_constructed = neightbour(a_vector, a_constructed)
+        b_value = f(os.path.join(out_dir, "selection", str(i).zfill(2)), config_dir, config_name, corpus_dir, folds, vector = b_vector, constructed = b_constructed)
+        prob = P(a_value, b_value, temperature)        
+        if b_value > a_value:
+            a_vector = b_vector
+            a_constructed = b_constructed
+        elif random.randint(0, 1) < prob:
+            a_vector = b_vector
+            a_constructed = b_constructed
+        temperature = temperature * 0.95
+        i += 1
               
 if __name__ == '__main__':
     go()