继续看yusugomori的代码，看逻辑回归。在DBN（Deep Blief Network）中，下面几层是RBM，最上层就是LR了。关于回归、二类回归、以及逻辑回归，资料就是前面转的几篇。套路就是设定目标函数（softmax损失函数），对参数求偏导数，得出权重更新公式等。

LogisticRegression.h注释如下：

class LogisticRegression

{

public:

  	int N;  		// number of input samples

  	int n_in;		// number of input nodes

  	int n_out;		// number of output nodes

  	double **W;		// weights connecting the input nodes and the output nodes

  	double *b;		// bias of the output nodes

  	// allocate memory and initialize the parameters

  	LogisticRegression(

  			int, 	// N

  			int, 	// n_in

  			int		// n_out

		  	);

  	~LogisticRegression();

public:

	// train the logistic regression model, update the value of W and b

  	void train (

	  		int*, 	// the input from input nodes in training set

	  		int*, 	// the output from output nodes in training set

	  		double	// the learning rate

		  );

 	// calculate the softmax for a input vector

 	// dSoftMax = exp(d_i - Max) / sum_i( exp(d_i - Max) )

  	void softmax (

	  		double*	// the calculated softmax probabiltiy -- input & output

			  );

	// do prediction by calculating the softmax probability from input

  	void predict (

	  		int*, 	// the input from input nodes in testing set

	  		double*	// the calculated softmax probability

			  );

};

顺便提一句。从前RBM的那个注释，是在家用VS2008写的；现在这个，用CFree5.0，轻量级、编辑器操作贴心，赞一下！