(self,features,labels)
| 90 | return C_w |
| 91 | |
| 92 | def train(self,features,labels): |
| 93 | |
| 94 | self._init_parameters_(features,labels) |
| 95 | |
| 96 | for times in xrange(self.M): |
| 97 | logging.debug('iterater %d' % times) |
| 98 | |
| 99 | |
| 100 | C_w = self.build_c_w() |
| 101 | min_error = ctypes.c_double(100000) |
| 102 | is_less = ctypes.c_int(-1) |
| 103 | feature_index = ctypes.c_int(-1) |
| 104 | |
| 105 | index = self.lib.find_min_error(self.X_matrix,self.n,self.N,self.C_Y,C_w,ctypes.byref(min_error),ctypes.byref(is_less),ctypes.byref(feature_index)) |
| 106 | |
| 107 | |
| 108 | |
| 109 | em = min_error.value |
| 110 | best_classifier = (em,feature_index.value,Sign(is_less.value,index)) #(误差率,针对的特征,分类器) |
| 111 | print 'em is %s, index is %s' % (str(em),str(feature_index.value)) |
| 112 | |
| 113 | |
| 114 | if em==0: |
| 115 | self.alpha.append(100) |
| 116 | else: |
| 117 | self.alpha.append(0.5*math.log((1-em)/em)) |
| 118 | |
| 119 | self.classifier.append(best_classifier[1:]) |
| 120 | |
| 121 | Z = self._Z_(best_classifier[1],best_classifier[2]) |
| 122 | |
| 123 | for i in xrange(self.N): |
| 124 | self.w[i] = self._w_(best_classifier[1],best_classifier[2],i)/Z |
| 125 | |
| 126 | def _predict_(self,feature): |
| 127 |
no test coverage detected