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

Adjust weight init methods.

Browse Source
This commit is contained in:
Marcus Edel
2015-04-25 19:55:23 +02:00
parent 8394ec0a6b
commit 2eb28bcc2a
7 changed files with 28 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/mlpack/methods/ann/connections/full_connection.hpp
+1 -1
View File
@@ -29,7 +29,7 @@ template<
typename InputLayerType,
typename OutputLayerType,
typename OptimizerType,
class WeightInitRule = NguyenWidrowInitialization<>,
class WeightInitRule = NguyenWidrowInitialization,
typename MatType = arma::mat,
typename VecType = arma::colvec
>
src/mlpack/methods/ann/connections/fullself_connection.hpp
+1 -1
View File
@@ -31,7 +31,7 @@ template<
typename InputLayerType,
typename OutputLayerType,
typename OptimizerType,
class WeightInitRule = NguyenWidrowInitialization<>,
class WeightInitRule = NguyenWidrowInitialization,
typename MatType = arma::mat,
typename VecType = arma::colvec
>
src/mlpack/methods/ann/connections/self_connection.hpp
+1 -1
View File
@@ -31,7 +31,7 @@ template<
typename InputLayerType,
typename OutputLayerType,
typename OptimizerType,
class WeightInitRule = NguyenWidrowInitialization<>,
class WeightInitRule = NguyenWidrowInitialization,
typename MatType = arma::mat,
typename VecType = arma::colvec
>
src/mlpack/methods/ann/layer/lstm_layer.hpp
+1 -1
View File
@@ -36,7 +36,7 @@ template <
class GateActivationFunction = LogisticFunction,
class StateActivationFunction = TanhFunction,
class OutputActivationFunction = TanhFunction,
class WeightInitRule = NguyenWidrowInitialization<>,
class WeightInitRule = NguyenWidrowInitialization,
typename OptimizerType = SteepestDescent<>,
typename MatType = arma::mat,
typename VecType = arma::colvec
src/mlpack/tests/feedforward_network_test.cpp
+12 -12
View File
@@ -159,12 +159,12 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkTest)
arma::mat testLabels = dataset.submat(dataset.n_rows - 3, 0,
dataset.n_rows - 1, dataset.n_cols - 1);
RandomInitialization<> randInitA(1, 2);
RandomInitialization randInitA(1, 2);
// Vanilla neural net with logistic activation function.
// Because 92 percent of the patients are not hyperthyroid a the neural
// network mst be significant better than 92%.
BuildVanillaNetwork<RandomInitialization<>,
BuildVanillaNetwork<RandomInitialization,
LogisticFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -174,7 +174,7 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkTest)
dataset.load("mnist_first250_training_4s_and_9s.arm");
RandomInitialization<> randInitB(-0.5, 0.5);
RandomInitialization randInitB(-0.5, 0.5);
// Normalize each point since these are images.
for (size_t i = 0; i < dataset.n_cols; ++i)
@@ -184,7 +184,7 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkTest)
labels.submat(0, labels.n_cols / 2, 0, labels.n_cols - 1).fill(1);
// Vanilla neural net with logistic activation function.
BuildVanillaNetwork<RandomInitialization<>,
BuildVanillaNetwork<RandomInitialization,
LogisticFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -192,7 +192,7 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkTest)
(dataset, labels, dataset, labels, 100, 100, 0.6, 10, randInitB);
// Vanilla neural net with tanh activation function.
BuildVanillaNetwork<RandomInitialization<>,
BuildVanillaNetwork<RandomInitialization,
TanhFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -208,7 +208,7 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkConvergenceTest)
arma::mat input;
arma::mat labels;
RandomInitialization<> randInit(0.5, 1);
RandomInitialization randInit(0.5, 1);
// Test on a non-linearly separable dataset (XOR).
input << 0 << 1 << 1 << 0 << arma::endr
@@ -216,7 +216,7 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkConvergenceTest)
labels << 0 << 0 << 1 << 1;
// Vanilla neural net with logistic activation function.
BuildVanillaNetwork<RandomInitialization<>,
BuildVanillaNetwork<RandomInitialization,
LogisticFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -224,7 +224,7 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkConvergenceTest)
(input, labels, input, labels, 4, 0, 0, 0.01, randInit);
// Vanilla neural net with tanh activation function.
BuildVanillaNetwork<RandomInitialization<>,
BuildVanillaNetwork<RandomInitialization,
TanhFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -237,7 +237,7 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkConvergenceTest)
labels << 0 << 0 << 1 << 0;
// vanilla neural net with sigmoid activation function.
BuildVanillaNetwork<RandomInitialization<>,
BuildVanillaNetwork<RandomInitialization,
LogisticFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -245,7 +245,7 @@ BOOST_AUTO_TEST_CASE(VanillaNetworkConvergenceTest)
(input, labels, input, labels, 4, 0, 0, 0.01, randInit);
// Vanilla neural net with tanh activation function.
BuildVanillaNetwork<RandomInitialization<>,
BuildVanillaNetwork<RandomInitialization,
TanhFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -358,7 +358,7 @@ BOOST_AUTO_TEST_CASE(NetworkDecreasingErrorTest)
arma::mat dataset;
dataset.load("mnist_first250_training_4s_and_9s.arm");
RandomInitialization<> randInitB(-0.5, 0.5);
RandomInitialization randInitB(-0.5, 0.5);
// Normalize each point since these are images.
for (size_t i = 0; i < dataset.n_cols; ++i)
@@ -368,7 +368,7 @@ BOOST_AUTO_TEST_CASE(NetworkDecreasingErrorTest)
labels.submat(0, labels.n_cols / 2, 0, labels.n_cols - 1) += 1;
// Vanilla neural net with logistic activation function.
BuildNetworkOptimzer<RandomInitialization<>,
BuildNetworkOptimzer<RandomInitialization,
LogisticFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
src/mlpack/tests/init_rules_test.cpp
+3 -3
View File
@@ -25,7 +25,7 @@ BOOST_AUTO_TEST_SUITE(InitRulesTest);
BOOST_AUTO_TEST_CASE(ConstantInitTest)
{
arma::mat weights;
RandomInitialization<> constantInit(1, 1);
RandomInitialization constantInit(1, 1);
constantInit.Initialize(weights, 100, 100);
bool b = arma::all(arma::vectorise(weights) == 1);
@@ -36,7 +36,7 @@ BOOST_AUTO_TEST_CASE(ConstantInitTest)
BOOST_AUTO_TEST_CASE(OrthogonalInitTest)
{
arma::mat weights;
OrthogonalInitialization<> orthogonalInit;
OrthogonalInitialization orthogonalInit;
orthogonalInit.Initialize(weights, 100, 200);
arma::mat orthogonalWeights = arma::eye<arma::mat>(100, 100);
@@ -60,7 +60,7 @@ BOOST_AUTO_TEST_CASE(OrthogonalInitGainTest)
arma::mat weights;
const double gain = 2;
OrthogonalInitialization<> orthogonalInit(gain);
OrthogonalInitialization orthogonalInit(gain);
orthogonalInit.Initialize(weights, 100, 200);
arma::mat orthogonalWeights = arma::eye<arma::mat>(100, 100);
src/mlpack/tests/recurrent_network_test.cpp
+9 -9
View File
@@ -119,7 +119,7 @@ BOOST_AUTO_TEST_CASE(SequenceClassificationTest)
SteepestDescent< > conOptimizer3(hiddenLayer0.InputSize(),
hiddenLayer1.OutputSize(), 1, 0);
NguyenWidrowInitialization<> randInit;
NguyenWidrowInitialization randInit;
FullConnection<
decltype(inputLayer),
@@ -316,7 +316,7 @@ BOOST_AUTO_TEST_CASE(FeedForwardRecurrentNetworkTest)
arma::mat input;
arma::mat labels;
RandomInitialization<> randInit(1, 1);
RandomInitialization randInit(1, 1);
// Test on a non-linearly separable dataset (XOR).
input << 0 << 1 << 1 << 0 << arma::endr
@@ -324,7 +324,7 @@ BOOST_AUTO_TEST_CASE(FeedForwardRecurrentNetworkTest)
labels << 0 << 0 << 1 << 1;
// Vanilla neural net with logistic activation function.
CompareVanillaNetworks<RandomInitialization<>,
CompareVanillaNetworks<RandomInitialization,
LogisticFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -332,7 +332,7 @@ BOOST_AUTO_TEST_CASE(FeedForwardRecurrentNetworkTest)
(input, labels, input, labels, 10, 10, randInit);
// Vanilla neural net with identity activation function.
CompareVanillaNetworks<RandomInitialization<>,
CompareVanillaNetworks<RandomInitialization,
IdentityFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -340,7 +340,7 @@ BOOST_AUTO_TEST_CASE(FeedForwardRecurrentNetworkTest)
(input, labels, input, labels, 1, 1, randInit);
// Vanilla neural net with rectifier activation function.
CompareVanillaNetworks<RandomInitialization<>,
CompareVanillaNetworks<RandomInitialization,
RectifierFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -348,7 +348,7 @@ BOOST_AUTO_TEST_CASE(FeedForwardRecurrentNetworkTest)
(input, labels, input, labels, 10, 10, randInit);
// Vanilla neural net with softsign activation function.
CompareVanillaNetworks<RandomInitialization<>,
CompareVanillaNetworks<RandomInitialization,
SoftsignFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -356,7 +356,7 @@ BOOST_AUTO_TEST_CASE(FeedForwardRecurrentNetworkTest)
(input, labels, input, labels, 10, 10, randInit);
// Vanilla neural net with tanh activation function.
CompareVanillaNetworks<RandomInitialization<>,
CompareVanillaNetworks<RandomInitialization,
TanhFunction,
SteepestDescent<>,
BinaryClassificationLayer<>,
@@ -585,7 +585,7 @@ void ReberGrammarTestNetwork(HiddenLayerType& hiddenLayer0,
SteepestDescent< > conOptimizer3(hiddenLayer0.OutputSize(),
hiddenLayer1.InputSize(), 0.1);
NguyenWidrowInitialization<> randInit;
NguyenWidrowInitialization randInit;
FullConnection<
decltype(inputLayer),
@@ -820,7 +820,7 @@ void DistractedSequenceRecallTestNetwork(HiddenLayerType& hiddenLayer0)
SteepestDescent< > conOptimizer3(hiddenLayer0.OutputSize(),
hiddenLayer1.InputSize(), 0.1);
NguyenWidrowInitialization<> randInit;
NguyenWidrowInitialization randInit;
FullConnection<
decltype(inputLayer),