Question

dlib 19.13-Linux Mint 18 g ++（Ubuntu 5.4.0-6ubuntu1〜16.04.9）5.4.0 20160609

尝试使用以下代码段与dlib进行XOR NN：

...

std::vector<matrix<unsigned long>> training_set;
std::vector<unsigned long>         training_labels;
std::vector<matrix<unsigned long>> testing_set;
std::vector<unsigned long>         testing_labels;

training_set.push_back( {0,0} );
training_set.push_back( {0,1} );
training_set.push_back( {1,0} );
training_set.push_back( {1,1} );

training_labels.push_back(0);
training_labels.push_back(1);
training_labels.push_back(1);
training_labels.push_back(0);


testing_set.push_back( {0,0} );
testing_set.push_back( {0,1} );
testing_set.push_back( {1,0} );
testing_set.push_back( {1,1} );

testing_labels.push_back(0);
testing_labels.push_back(1);
testing_labels.push_back(1);
testing_labels.push_back(0);


using net_type = loss_multiclass_log<
                            fc<1,        
                            relu<fc<3,   
                            input<matrix<unsigned long>> 
                            >>>>;

...

它可以编译，但是我在执行测试时得到以下输出：

在第322行检测到错误。在文件../dlib/dnn/loss.h中检测到错误。在函数double dlib :: loss_multiclass_log _ :: compute_loss_value_and_gradient（const dlib :: tensor＆，const_label_iterator，SUBNET＆）中检测到错误const [with const_label_iterator = __gnu_cxx :: __ normal_iterator>; SUBNET = dlib :: dimpl :: subnet_wrapper，dlib :: add_layer，dlib :: input>，void>，void>，void>，true，void>]。

失败的表达式是y

Answer 1

我认为dlib的fc类的第一个参数是输出节点的数量。但事实证明，这似乎是可能的产出数量。因此0和1是该层的两个可能的输出值。从1更改为2，然后重新编译网络火车，并显示出100％的准确性：

using net_type = loss_multiclass_log<
                            fc<2,  // <--- changed to 2    
                            relu<fc<3,   
                            input<matrix<unsigned long>> 
                            >>>>;

dlib：失败的表达式为y <output_tensor.k（）

1 个答案: