apaam/mlpack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix style, Ajust to one line,

Browse Source 
Signed-off-by: Omar Shrit <omar@avontech.fr>
This commit is contained in:
Omar Shrit
2024-01-11 11:56:52 +01:00
parent cb5ac5be3c
commit a101b0801c
1 changed files with 6 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/mlpack/methods/ann/layer/batch_norm_impl.hpp
+6 -12
View File
@@ -216,8 +216,7 @@ void BatchNormType<MatType>::Forward(
// Calculate mean and variance over all channels.
MatType mean = arma::sum(arma::sum(inputTemp, 2), 0) / m;
variance = arma::sum(arma::sum(arma::pow(
inputTemp.each_slice() - repmat(mean,
inputSize, 1), 2), 2), 0) / m;
inputTemp.each_slice() - repmat(mean, inputSize, 1), 2), 2), 0) / m;
outputTemp.each_slice() -= repmat(mean, inputSize, 1);
@@ -226,8 +225,7 @@ void BatchNormType<MatType>::Forward(
inputMean = outputTemp;
// Normalize output.
outputTemp.each_slice() /= arma::sqrt(repmat(variance,
inputSize, 1) + eps);
outputTemp.each_slice() /= arma::sqrt(repmat(variance, inputSize, 1) + eps);
// Re-used in backward propagation.
normalized.set_size(arma::size(inputTemp));
@@ -260,14 +258,11 @@ void BatchNormType<MatType>::Forward(
const_cast<MatType&>(output).memptr(), inputSize, size,
batchSize * higherDimension, false, false);
outputTemp.each_slice() -= repmat(runningMean.t(),
inputSize, 1);
outputTemp.each_slice() -= repmat(runningMean.t(), inputSize, 1);
outputTemp.each_slice() /= arma::sqrt(repmat(runningVariance.t(),
inputSize, 1) + eps);
outputTemp.each_slice() %= repmat(gamma.t(),
inputSize, 1);
outputTemp.each_slice() += repmat(beta.t(),
inputSize, 1);
outputTemp.each_slice() %= repmat(gamma.t(), inputSize, 1);
outputTemp.each_slice() += repmat(beta.t(), inputSize, 1);
}
}
@@ -301,8 +296,7 @@ void BatchNormType<MatType>::Backward(
// Step 3: dl / dxhat * 1 / stdInv + variance * 2 * (x - mu) / m +
// dl / dmu * 1 / m.
gTemp = (norm.each_slice() % repmat(stdInv,
inputSize, 1)) +
gTemp = (norm.each_slice() % repmat(stdInv, inputSize, 1)) +
((inputMean.each_slice() % repmat(vars, inputSize, 1) * 2.0) / m);
// Step 4: sum (dl / dxhat * -1 / stdInv) + variance *