shashadehuajiang / trace_classifier Goto Github PK

   def data_enhancement(self,X):
        if self.cfg.PACKET2FLOW == '2dCNN':
            return X # not ready yet

        if self.cfg.PACKET2FLOW == '1dCNN' and self.cfg.SCALE_1dCNN:
            return X # not ready yet

        X_new = copy.deepcopy(X)
        
        for i_f in range(len(X_new)):
            for i_v in range(len(X_new[i_f])):
                X_new[i_f][i_v] = X_new[i_f][i_v][0:2]

        rand_num = random.random()

        if rand_num>0.5 and rand_num<0.7:
            # 随机丢弃
            for i_f in range(len(X_new)):
                for i_v in range(len(X_new[i_f])-1,-1,-1):
                    rand_threshold = random.random()*0.1
                    if random.random() < rand_threshold:
                        del X_new[i_f][i_v]

        if rand_num>0.7 and rand_num<0.8:
            # 时间平移（保序）
            for i_f in range(len(X_new)):
                for i_v in range(1,len(X_new[i_f])):
                    if i_v == len(X_new[i_f]) - 1:
                        max_size = X_new[i_f][i_v][0]/2
                        random_biaos = (random.random()*2-1)*max_size
                        X_new[i_f][i_v][0] -= random_biaos
                    else:
                        max_size = min(X_new[i_f][i_v][0],X_new[i_f][i_v+1][0])
                        random_biaos = (random.random()*2-1)*max_size
                        X_new[i_f][i_v][0] -= random_biaos
                        X_new[i_f][i_v+1][0] += random_biaos
        
        if rand_num>0.8 and rand_num<1:
            # 前后随机cut
            max_time_list = [max([X_new[i_f][i_v][0] for i_v in range(0,len(X_new[i_f]))]) for i_f in range(len(X_new))] 
            max_time = max(max_time_list)
            
            startcut = (random.random()*0.2) * max_time
            endcut = (1 - random.random()*0.2) * max_time
            for i_f in range(len(X_new)-1,-1,-1):
                for i_v in range(len(X_new[i_f])-1,-1,-1):
                    if X_new[i_f][i_v][0] < startcut or X_new[i_f][i_v][0] > endcut:
                        del X_new[i_f][i_v]
                if len(X_new[i_f]) == 0:
                    del X_new[i_f]



        for i_f in range(len(X_new)):
            for i_v in range(len(X_new[i_f])):
                # 添加绝对时间
                if i_v>=1:
                    time_sum = X_new[i_f][i_v][0] + X_new[i_f][i_v-1][-1]
                    X_new[i_f][i_v] = np.r_[X_new[i_f][i_v],time_sum]
                else:
                    X_new[i_f][i_v] = np.r_[X_new[i_f][i_v],0]
                # 添加顺序，并clip
                id_pad = min(i_v/100,1)
                X_new[i_f][i_v] = np.r_[X_new[i_f][i_v],id_pad]
                # 添加百分比
                percentage = i_v/len(X_new[i_f])
                X_new[i_f][i_v] = np.r_[X_new[i_f][i_v],percentage]

        return X