zero to one

README.md

@@ -64,6 +64,33 @@ make install
 
 ## 新的demo程序实现5层全连接层,可自定义神经元和激活函数,损失函数
 
+全连接层使用方法:
+
+第一层的权重自定义,而后调用forward函数前向传播一层,自动求出激活以后的值,激活函数可自定义.
+
+首先定义一个权重矩阵和偏置矩阵,第一个矩阵的维度大小使用数据列去定义:
+
+```c
+	Matrix bias1 = CreateRandMat(2,1);
+	Matrix weight1 = CreateRandMat(2,data.col);
+```
+
+之后可以输出第一层前向传播的值,同时可以定义下一层的bias的维度, row使用第一层的权重矩阵的行,第二层的权重矩阵的行使用了第一层的输出的行, 而列自行定义即可, 这一点体现了前向传播算法的维度相容. 也就是:
+
+```c
+Matrix output1 = sequaltial.forward(get_T(get_row(data_mine,index)),weight1,bias1);
+```
+
+```c
+Matrix weight2 = CreateRandMat(output1.row,2);
+Matrix bias2 = CreateRandMat(weight2.row,1);
+Matrix output2 = sequaltial.forward(output1,weight2,bias2);
+```
+
+同时第二层的输出也可以求出来,以此类推 .
+
+最终输出代码见nerual_test.cpp ![nerual_test1](/Users/zhangyiji/Documents/code/cpp_demo/my_os/Edge-Computing-Engine/picture/nerual_test1.png)
+
 代码:
 
 ```c

grad_edge/matrix_grad.h

@@ -18,9 +18,6 @@ Node sigmoid(Node z){
 struct edge_network
 {
 	edge_network(int input, int num_neuron){			
-
-
-
     }
 	Matrix forward(Matrix data,Matrix weights,Matrix bais)
 	{

main.cpp

@@ -5,7 +5,7 @@
 #include <math.h>
 #include <fstream>
 #include "./autodiff/node.h"
-#include"./matrix/matrix_def.h"a
+#include"./matrix/matrix_def.h"
 #include"./matrix/matrix_pro.h"
 #include"./welcome/score_wel.cpp"
 #include"./logistic/logistic_def.h"
@@ -29,38 +29,7 @@ return sigmoid_act;
 Node (*loss)(Node,Node) = loss_act;
 Node (*act)(Node) = sigmoid_act;
 int main()
-{	 	/*
-	   welcome();	
-	   string path = "./data/new_data2.csv";
-	Matrix data = read_csv(path);
-	Matrix bais = CreateMatrix(data.row,1);		
-	data = appply(data,bais,1);
-	Matrix y = iloc(data,0,0,3,4);
-	Matrix x_1 = iloc(data,0,0,0,3);
-	Matrix x_2 = get_T(x_1);
-	double alpha = 0.002;
-	int max_epoch = 1;
-	Matrix weight = CreateMatrix(3,1);
-	change_va(weight,0,0,1);
-	change_va(weight,1,0,1);
-	change_va(weight,2,0,1);
-	int epoch = 0;
-	for(epoch = 0;epoch<=max_epoch;epoch++)
-	{
-	cout<<"-----------split-line-----------"<<endl;			
-	Matrix temp_mul = mul(x_1,weight);
-	Matrix h =e_sigmoid(temp_mul);
-	Matrix error = subtract(y,h);
-	Matrix temp_update = mul(x_2,error);
-	Matrix updata = add(weight,times_mat(alpha,temp_update),0);
-	cout_mat(weight);
-	cout<<"epoch: "<<epoch<<" error: "<<matrix_sum(error)<<endl;
-	cout<<"-----------split-line-----------"<<endl;	
-	}
-	stop = clock();
-    printf("%f\n", (double)(stop - start) / CLOCKS_PER_SEC);
-	*/
-	
+{
 	cout<<"------------autodiff for neraul network-----------"<<endl;
 	Matrix data_mine = CreateMatrix(2,1);
 	change_va(data_mine,0,0,0.55);
@@ -107,30 +76,7 @@ int main()
 	bais1 = subtract(bais1,times_mat(0.001,backward3));
 	weight2 = subtract(weight2,times_mat(0.001,weight_2_grad));
 	bais2 = subtract(bais2,times_mat(0.001,output_end));
-	map<string,int>count_word;
-	string path_word = "mytest.csv";
-	str_Matrix data_readed = read_file(path_word);
-	for(int index_x = 0;index_x<data_readed.row;index_x++)
-	{
-		for(int index_y = 0;index_y<data_readed.col;index_y++)
-		{
-		string word = data_readed.str_matrix[index_x][index_y];
-			if(count_word.count(data_readed.str_matrix[index_x][index_y]) == 0)
-			{
-				count_word[word] = 1;
-			}
-			else
-			{
-				count_word[word]=count_word[word]+1;
-			}
-		}
-
-	}
-		cout<<"class: "<<count_word["class"]<<endl;
-		cout<<"classes: "<<count_word["classes"]<<endl;
-		cout<<"classification: "<<count_word["classification"]<<endl;
-		cout<<"classifiers: "<<count_word["classifiers"]<<endl;
+	cout<<"neraul end;"<<endl;
 	return 0;
-
 	}
 }

matrix/matrix_pro.h

@@ -331,4 +331,14 @@ Matrix padding(Matrix mid1,int shape1,int shape2)
 	}
 	return result;
 }
+Matrix get_row(Matrix mid1,int index)
+{
+	int index_y = 0;
+	Matrix mid2 = CreateMatrix(1,mid1.col);
+	for(index_y=0;index_y<mid1.col;index_y++)
+	{
+	mid2.matrix[0][index_y] = mid1.matrix[index][index_y];
+	}
+	return mid2;
+}
 #endif

nerual_test.cpp

+#include<iostream>
+#include<time.h>
+#include<string>
+#include<math.h>
+#include<fstream>
+#include"./autodiff/node.h"
+#include"./matrix/matrix_def.h"
+#include"./matrix/matrix_pro.h"
+#include"./welcome/score_wel.cpp"
+#include"./logistic/logistic_def.h"
+#include"./file_pro/data_read.h"
+#include"./grad_edge/matrix_grad.h"
+using namespace std;
+Node z = 1;
+Node t1 = 1,a13 = 1;
+Node loss_act(Node t1,Node a13)
+{
+	Node loss = 0.5*(pow((t1-a13),2));
+	return loss;
+}
+Node sigmoid_act(Node z)
+{
+	Node sigmoid_act = 1/(1+(1/exp(z)));
+	return sigmoid_act;
+}
+Node (*loss)(Node,Node) = loss_act;
+Node (*act)(Node) = sigmoid_act;
+int main()
+{
+	cout<<"begin to train"<<endl;
+	int input_dim = 2;
+	int output_dim = 2;
+	edge_network sequaltial(input_dim,output_dim);
+	Matrix data_mine = CreateMatrix(100,2);
+	Matrix label_mine = CreateRandMat(100,1);
+	Matrix data = iloc(data_mine,0,1,0,0);
+	cout<<"data: ";
+	cout_mat(data);
+	Matrix label = iloc(label_mine,0,1,0,0);
+	cout<<"label: ";
+	cout_mat(label);
+	Matrix bias1 = CreateRandMat(2,1);
+	Matrix weight1 = CreateRandMat(2,data.col);
+	cout<<"weight1: ";
+	cout_mat(weight1);
+	int index = 0;
+	Matrix output1 = sequaltial.forward(get_T(get_row(data_mine,index)),weight1,bias1);
+	Matrix weight2 = CreateRandMat(output1.row,2);
+	cout<<"weight2: "<<endl;
+	cout_mat(weight2);
+	Matrix bias2 = CreateRandMat(weight2.row,1);
+	Matrix output2 = sequaltial.forward(output1,weight2,bias2);
+	cout_mat(output2);
+	cout<<"output1:"<<endl;
+	cout_mat(output1);
+
+}
+