【deep learning学习笔记】注释yusugomori的DA代码 --- dA.h

DA就是“Denoising Autoencoders”的缩写。继续给yusugomori做注释，边注释边学习。看了一些DA的材料，基本上都在前面“转载”了。学习中间总有个疑问：DA和RBM到底啥区别？（别笑，我不是“学院派”的看Deep Learning理论，如果“顺次”看下来，可能不会有这个问题），现在了解的差不多了，详情见：【deep learning学习笔记】Autoencoder。之后，又有个疑问，DA具体的权重更新公式是怎么推导出来的？我知道是BP算法，不过具体公示的推导、偏导数的求解，没有看到哪个材料有具体的公式，所以姑且认为yusugomori的代码写的是正确的。

注释后的头文件：

// The Class of denoising auto-encoderclass dA {public:int N;// the number of training samplesint n_visible;// the number of visible nodesint n_hidden;// the number of hidden nodesdouble **W;// the weight connecting visible node and hidden nodedouble *hbias;// the bias of hidden nodesdouble *vbias;// the bias of visible nodespublic:// initialize the parametersdA ( int,// N int,// n_visible int ,// n_hidden double**,// W double*,// hbias double*// vbias );~dA();// make the input noisedvoid get_corrupted_input (int*,// the original input 0-1 vector-- inputint*,// the resulted 0-1 vector gotten noised-- outputdouble// the p probability of noise, binomial test -- input);// encode process: calculate the probability output from hidden node// p(hi|v) = sigmod ( sum_j(vj * wij) + bi), it's same with RBM// but different from RBM, it dose not generate 0-1 state from Bernoulli distributionvoid get_hidden_values (int*,// the input from visible nodesdouble*// the output of hidden nodes);// decode process: calculate the probability output from visiable node// p(vi|h) = sigmod ( sum_j(hj * wij) + ci), it's same with RBM// but different from RBM, it dose not generate 0-1 state from Bernoulli distribution void get_reconstructed_input (double*,// the input from hidden nodesdouble*// the output reconstructed of visible nodes);// train the model by a single samplevoid train (int*,// the input sample from visiable nodedouble,// the learning ratedouble// corruption_level is the probability of noise);// reconstruct the input samplevoid reconstruct (int*,// the input sample-- inputdouble*// the reconstructed value -- output);};