矩阵乘法的结果是 'nan'
Result of matrix-multiplication is 'nan'
我的代码中有两个矩阵乘法。第一个工作正常,但第二个给了我 'nan' 输出。
乙状结肠:
double sigmoide(double value) {
return 1.0 / (1.0 + exp(-value));
}
初始化:
double LO = -0.5;
double HI = 0.5;
double input[50][2];
double hiddenWeights[2][10];
double outputWeights[11][1];
double hiddenResults[50][11];
double outputResults[50][1];
for (int i = 0; i < 50; i++) {
input[i][0] = 1.0f; /// Bias
input[i][1] = (7.0f/50.0f) * (double)i; /// Samples
}
for (int i = 0; i < 50; i++) {
outputSetpoint[i][0] = sin((7.0f/50.0f) * (double)i);
}
-0.5 到 0.5 之间的随机值:
for (int i = 0; i < 2; i++) {
for (int j = 0; j < 10; j++) {
hiddenWeights[i][j] = LO + static_cast <float> (rand()) /( static_cast <float> (RAND_MAX/(HI-LO)));
}
}
for (int i = 0; i < 11; i++) {
for (int j = 0; j < 1; j++) {
outputWeights[i][j] = LO + static_cast <float> (rand()) /( static_cast <float> (RAND_MAX/(HI-LO)));
}
}
矩阵乘法:
for (int s = 0; s < 50; s++) {
for (int j = 0; j < 10; j++) {
for (int i = 0; i < 2; i++) {
// First matrix-multiplication
hiddenResults[s][j] += nexttowardf(input[s][i] * hiddenWeights[i][j], 0.0f);
}
hiddenResults[s][10] = 1.0f;
hiddenResults[s][j] = sigmoide(hiddenResults[s][j]);
}
for (int j = 0; j < 1; j++) {
for (int i = 0; i < 11; i++) {
// Second matrix-multiplication
outputResults[s][j] += hiddenResults[s][i] * outputWeights[i][j];
}
outputResults[s][j] = sigmoide(outputResults[s][j]);
error = outputSetpoint[s][j] - outputResults[s][j];
}
std::cout << outputResults[s][0] << std::endl;
}
输出:
nan
1
0.287491
0.432262
0.293168
0.336324
0.283587
0.282668
1
0.333261
nan
0.279217
nan
0.239026
0
0.338551
0.274522
0.209411
0.24247
0.273364
0.179109
0.199499
0.271506
1
nan
nan
nan
nan
您忘记初始化 hiddenResults
和 outputResults
这样做:
double hiddenResults[50][11] = {0};
double outputResults[50][1] = {0};
一边初始化其他变量。
养成好习惯。
我的代码中有两个矩阵乘法。第一个工作正常,但第二个给了我 'nan' 输出。
乙状结肠:
double sigmoide(double value) {
return 1.0 / (1.0 + exp(-value));
}
初始化:
double LO = -0.5;
double HI = 0.5;
double input[50][2];
double hiddenWeights[2][10];
double outputWeights[11][1];
double hiddenResults[50][11];
double outputResults[50][1];
for (int i = 0; i < 50; i++) {
input[i][0] = 1.0f; /// Bias
input[i][1] = (7.0f/50.0f) * (double)i; /// Samples
}
for (int i = 0; i < 50; i++) {
outputSetpoint[i][0] = sin((7.0f/50.0f) * (double)i);
}
-0.5 到 0.5 之间的随机值:
for (int i = 0; i < 2; i++) {
for (int j = 0; j < 10; j++) {
hiddenWeights[i][j] = LO + static_cast <float> (rand()) /( static_cast <float> (RAND_MAX/(HI-LO)));
}
}
for (int i = 0; i < 11; i++) {
for (int j = 0; j < 1; j++) {
outputWeights[i][j] = LO + static_cast <float> (rand()) /( static_cast <float> (RAND_MAX/(HI-LO)));
}
}
矩阵乘法:
for (int s = 0; s < 50; s++) {
for (int j = 0; j < 10; j++) {
for (int i = 0; i < 2; i++) {
// First matrix-multiplication
hiddenResults[s][j] += nexttowardf(input[s][i] * hiddenWeights[i][j], 0.0f);
}
hiddenResults[s][10] = 1.0f;
hiddenResults[s][j] = sigmoide(hiddenResults[s][j]);
}
for (int j = 0; j < 1; j++) {
for (int i = 0; i < 11; i++) {
// Second matrix-multiplication
outputResults[s][j] += hiddenResults[s][i] * outputWeights[i][j];
}
outputResults[s][j] = sigmoide(outputResults[s][j]);
error = outputSetpoint[s][j] - outputResults[s][j];
}
std::cout << outputResults[s][0] << std::endl;
}
输出:
nan
1
0.287491
0.432262
0.293168
0.336324
0.283587
0.282668
1
0.333261
nan
0.279217
nan
0.239026
0
0.338551
0.274522
0.209411
0.24247
0.273364
0.179109
0.199499
0.271506
1
nan
nan
nan
nan
您忘记初始化 hiddenResults
和 outputResults
这样做:
double hiddenResults[50][11] = {0};
double outputResults[50][1] = {0};
一边初始化其他变量。
养成好习惯。