Perceptron演算法—C語言

本篇介紹第三個演算法，感知機演算法，同樣僅介紹兩個主要函式：預測函式和訓練函式。

需要說明，本演算法仍使用隨機梯度下降演算法，與上篇多元線性迴歸有很多相似之處。

演算法介紹

啟用函式可以如下表示：
$$activation=bias+\sum _{i=1}^{n}weigh{t}_i\times x_i$$

階躍函式
$$prediction=1.0IFactivation\ge 0.0ELSE0.0$$

函式

讀取csv

• 以下三個函式分別為獲取行數、獲取列數、獲取文字內容。

double **dataset;
int row,col;

int get_row(char *filename)//獲取行數
{
	char line[1024];
	int i = 0;
	FILE* stream = fopen(filename, "r");
	while(fgets(line, 1024, stream)){
		i++;
	}
	fclose(stream);
	return i;
}

int get_col(char *filename)//獲取列數
{
	char line[1024];
	int i = 0;
	FILE* stream = fopen(filename, "r");
	fgets(line, 1024, stream);
	char* token = strtok(line, ",");
	while(token){
		token = strtok(NULL, ",");
		i++;
	}
	fclose(stream);
	return i;
}

void get_two_dimension(char* line, double** data, char *filename)
{
	FILE* stream = fopen(filename, "r");
	int i = 0;
	while (fgets(line, 1024, stream))//逐行讀取
    {
    	int j = 0;
    	char *tok;
        char* tmp = strdup(line);
        for (tok = strtok(line, ","); tok && *tok; j++, tok = strtok(NULL, ",\n")){
        	data[i][j] = atof(tok);//轉換成浮點數
		}//字串拆分操作
        i++;
        free(tmp);
    }
    fclose(stream);//檔案開啟後要進行關閉操作
}

EXAMPLE

int main()
{
	char filename[] = "data.csv";
    char line[1024];
    double **data;
    int row, col;
    row = get_row(filename);
    col = get_col(filename);
    data = (double **)malloc(row * sizeof(int *));
	for (int i = 0; i < row; ++i){
		data[i] = (double *)malloc(col * sizeof(double));
	}//動態申請二維陣列
	get_two_dimension(line, data, filename);
	printf("row = %d\n", row);
	printf("col = %d\n", col);

	int i, j;
	for(i=0; i<row; i++){
		for(j=0; j<col; j++){

	float mean = (float)(sum / length);
	return mean;
}

隨機梯度下降估計迴歸係數

更新方程：
KaTeX parse error: Undefined control sequence: \ at position 29: …iction-expected\̲ ̲

$$b_1(t+1)=b_1(t)-learningrate\times error(t)\times x_1(t)$$

$$b_0(t+1)=b_0(t)-learningrate\times error(t)$$

// 引數為：資料集、每個資料集屬性個數(帶label)、權重、學習率、epoch、train_size
void train_weights(double **data, int col,double *weights, double l_rate, int n_epoch, int train_size) {
	int i;
	for (i = 0; i < n_epoch; i++) {
		int j = 0;//遍歷每一行
		for (j = 0; j < train_size; j++) {
			double yhat = predict(col,data[j], weights);
			double err = data[j][col - 1] - yhat;
			weights[0] += l_rate * err;
			int k;
			for (k = 0; k < col - 1; k++) {
				weights[k + 1] += l_rate * err * data[j][k];
			}
		}
	}
	for (i = 0; i < col; i++) {
		printf("weights[%d]=%f\n",i, weights[i]);
	}
}

預測函式

// 引數：樣本屬性個數、樣本、權重
double predict(int col,double *array, double *weights) {//預測某一行的值
	double activation = weights[0];
	int i;
	for (i = 0; i < col - 1; i++)
		activation += weights[i + 1] * array[i];
	double output = 0.0;
	if (activation >= 0.0)
		output = 1.0;
	else
		output = 0.0;
	return output;
}