[mlpack-git] master: Add implementation of the full connection class. (f362f79)

[gitdub at big.cc.gt.atl.ga.us]

Repository : https://github.com/mlpack/mlpack

On branch  : master
Link       : https://github.com/mlpack/mlpack/compare/904762495c039e345beba14c1142fd719b3bd50e...f94823c800ad6f7266995c700b1b630d5ffdcf40

>---------------------------------------------------------------

commit f362f7961acd98be9922ae7335fbe3e3c3af5322
Author: Marcus Edel <marcus.edel at fu-berlin.de>
Date:   Wed Dec 31 21:58:19 2014 +0100

    Add implementation of the full connection class.


>---------------------------------------------------------------

f362f7961acd98be9922ae7335fbe3e3c3af5322
 .../methods/ann/connections/full_connection.hpp    | 126 ++++++++++++++++++
 .../ann/connections/fullself_connection.hpp        | 147 +++++++++++++++++++++
 2 files changed, 273 insertions(+)

diff --git a/src/mlpack/methods/ann/connections/full_connection.hpp b/src/mlpack/methods/ann/connections/full_connection.hpp
new file mode 100644
index 0000000..82efd49
--- /dev/null
+++ b/src/mlpack/methods/ann/connections/full_connection.hpp
@@ -0,0 +1,126 @@
+/**
+ * @file full_connection.hpp
+ * @author Marcus Edel
+ *
+ * Implementation of the full connection class.
+ */
+#ifndef __MLPACK_METHOS_ANN_CONNECTIONS_FULL_CONNECTION_HPP
+#define __MLPACK_METHOS_ANN_CONNECTIONS_FULL_CONNECTION_HPP
+
+#include <mlpack/core.hpp>
+#include <mlpack/methods/ann/init_rules/nguyen_widrow_init.hpp>
+
+namespace mlpack {
+namespace ann /** Artificial Neural Network. */ {
+
+/**
+ * Implementation of the full connection class. The full connection connects
+ * every neuron from the input layer with the output layer in a matrix
+ * multiplicative way.
+ *
+ * @tparam InputLayerType Type of the connected input layer.
+ * @tparam OutputLayerType Type of the connected output layer.
+ * @tparam OptimizerType Type of the optimizer used to update the weights.
+ * @tparam WeightInitRule Rule used to initialize the weight matrix.
+ * @tparam MatType Type of data (arma::mat or arma::sp_mat).
+ * @tparam VecType Type of data (arma::colvec, arma::mat or arma::sp_mat).
+ */
+template<
+    typename InputLayerType,
+    typename OutputLayerType,
+    typename OptimizerType,
+    class WeightInitRule = NguyenWidrowInitialization<>,
+    typename MatType = arma::mat,
+    typename VecType = arma::colvec
+>
+class FullConnection
+{
+ public:
+  /**
+   * Create the FullConnection object using the specified input layer, output
+   * layer, optimizer and weight initialize rule.
+   *
+   * @param lowerBound The number used as lower bound.
+   * @param upperBound The number used as upper bound.
+   */
+  FullConnection(InputLayerType& inputLayer,
+                 OutputLayerType& outputLayer,
+                 OptimizerType& optimizer,
+                 WeightInitRule weightInitRule = WeightInitRule()) :
+      inputLayer(inputLayer), outputLayer(outputLayer), optimizer(optimizer)
+  {
+    weightInitRule.Initialize(weights, outputLayer.InputSize(),
+        inputLayer.OutputSize());
+  }
+
+  /**
+   * Ordinary feed forward pass of a neural network, evaluating the function
+   * f(x) by propagating the activity forward through f.
+   *
+   * @param input Input data used for evaluating the specified activity function.
+   */
+  void FeedForward(const VecType& input)
+  {
+    outputLayer.InputActivation() += (weights * input);
+  }
+
+  /**
+   * Ordinary feed backward pass of a neural network, calculating the function
+   * f(x) by propagating x backwards trough f. Using the results from the feed
+   * forward pass.
+   *
+   * @param error The backpropagated error.
+   */
+  void FeedBackward(const VecType& error)
+  {
+    // Calculating the delta using the partial derivative of the error with
+    // respect to a weight.
+    delta = (weights.t() * error);
+  }
+
+  //! Get the weights.
+  MatType& Weights() const { return weights; }
+  //! Modify the weights.
+  MatType& Weights() { return weights; }
+
+  //! Get the input layer.
+  InputLayerType& InputLayer() const { return inputLayer; }
+  //! Modify the input layer.
+  InputLayerType& InputLayer() { return inputLayer; }
+
+  //! Get the output layer.
+  OutputLayerType& OutputLayer() const { return outputLayer; }
+  //! Modify the output layer.
+  OutputLayerType& OutputLayer() { return outputLayer; }
+
+  //! Get the optimzer.
+  OptimizerType& Optimzer() const { return optimizer; }
+  //! Modify the optimzer.
+  OptimizerType& Optimzer() { return optimizer; }
+
+  //! Get the detla.
+  VecType& Delta() const { return delta; }
+ //  //! Modify the delta.
+  VecType& Delta() { return delta; }
+
+ private:
+  //! Locally-stored weight object.
+  MatType weights;
+
+  //! Locally-stored connected input layer object.
+  InputLayerType& inputLayer;
+
+  //! Locally-stored connected output layer object.
+  OutputLayerType& outputLayer;
+
+  //! Locally-stored optimzer object.
+  OptimizerType& optimizer;
+
+  //! Locally-stored detla object that holds the calculated delta.
+  VecType delta;
+}; // class FullConnection
+
+}; // namespace ann
+}; // namespace mlpack
+
+#endif
diff --git a/src/mlpack/methods/ann/connections/fullself_connection.hpp b/src/mlpack/methods/ann/connections/fullself_connection.hpp
new file mode 100644
index 0000000..5d7fc81
--- /dev/null
+++ b/src/mlpack/methods/ann/connections/fullself_connection.hpp
@@ -0,0 +1,147 @@
+/**
+ * @file fullself_connection.hpp
+ * @author Marcus Edel
+ *
+ * Implementation of the full self connection class. This is used mainly used as
+ * recurrent connection.
+ */
+#ifndef __MLPACK_METHOS_ANN_CONNECTIONS_FULLSELF_CONNECTION_HPP
+#define __MLPACK_METHOS_ANN_CONNECTIONS_FULLSELF_CONNECTION_HPP
+
+#include <mlpack/core.hpp>
+#include <mlpack/methods/ann/init_rules/nguyen_widrow_init.hpp>
+
+namespace mlpack {
+namespace ann /** Artificial Neural Network. */ {
+
+/**
+ * Implementation of the full self connection class. The full connection
+ * connects every neuron from the input layer with the output layer in a matrix
+ * multiplicative way. This connection is used as recurrent connection, the
+ * input would be automatically updated every time step.
+ *
+ * @tparam InputLayerType Type of the connected input layer.
+ * @tparam OutputLayerType Type of the connected output layer.
+ * @tparam OptimizerType Type of the optimizer used to update the weights.
+ * @tparam WeightInitRule Rule used to initialize the weight matrix.
+ * @tparam MatType Type of data (arma::mat or arma::sp_mat).
+ * @tparam VecType Type of data (arma::colvec, arma::mat or arma::sp_mat).
+ */
+template<
+    typename InputLayerType,
+    typename OutputLayerType,
+    typename OptimizerType,
+    class WeightInitRule = NguyenWidrowInitialization<>,
+    typename MatType = arma::mat,
+    typename VecType = arma::colvec
+>
+class FullselfConnection
+{
+ public:
+  /**
+   * Create the FullselfConnection object using the specified input layer,
+   * output layer, optimizer and weight initialize rule.
+   *
+   * @param lowerBound The number used as lower bound.
+   * @param upperBound The number used as upper bound.
+   */
+  FullselfConnection(InputLayerType& inputLayer,
+                  OutputLayerType& outputLayer,
+                  OptimizerType& optimizer,
+                  WeightInitRule weightInitRule = WeightInitRule()) :
+      inputLayer(inputLayer), outputLayer(outputLayer), optimizer(optimizer)
+  {
+    weightInitRule.Initialize(weights, outputLayer.InputSize(),
+        inputLayer.OutputSize());
+  }
+
+  /**
+   * Ordinary feed forward pass of a neural network, evaluating the function
+   * f(x) by propagating the activity forward through f.
+   *
+   * @param input Input data used for evaluating the specified activity function.
+   */
+  void FeedForward(const VecType& input)
+  {
+    outputLayer.InputActivation() += (weights * input);
+  }
+
+  /**
+   * Ordinary feed backward pass of a neural network, calculating the function
+   * f(x) by propagating x backwards trough f. Using the results from the feed
+   * forward pass.
+   *
+   * @param error The backpropagated error.
+   */
+  void FeedBackward(const VecType& error)
+  {
+    // Calculating the delta using the partial derivative of the error with
+    // respect to a weight.
+    delta = (weights.t() * error);
+  }
+
+  //! Get the weights.
+  MatType& Weights() const { return weights; }
+  //! Modify the weights.
+  MatType& Weights() { return weights; }
+
+  //! Get the input layer.
+  InputLayerType& InputLayer() const { return inputLayer; }
+  //! Modify the input layer.
+  InputLayerType& InputLayer() { return inputLayer; }
+
+  //! Get the output layer.
+  OutputLayerType& OutputLayer() const { return outputLayer; }
+  //! Modify the output layer.
+  OutputLayerType& OutputLayer() { return outputLayer; }
+
+  //! Get the optimzer.
+  OptimizerType& Optimzer() const { return optimizer; }
+  //! Modify the optimzer.
+  OptimizerType& Optimzer() { return optimizer; }
+
+  //! Get the detla.
+  VecType& Delta() const { return delta; }
+ //  //! Modify the delta.
+  VecType& Delta() { return delta; }
+
+ private:
+  //! Locally-stored weight object.
+  MatType weights;
+
+  //! Locally-stored connected input layer object.
+  InputLayerType& inputLayer;
+
+  //! Locally-stored connected output layer object.
+  OutputLayerType& outputLayer;
+
+  //! Locally-stored optimzer object.
+  OptimizerType& optimizer;
+
+  //! Locally-stored detla object that holds the calculated delta.
+  VecType delta;
+}; // class RandomInitialization
+
+
+//! Connection traits for the full self connection.
+template<
+    typename InputLayerType,
+    typename OutputLayerType,
+    typename OptimizerType,
+    class WeightInitRule,
+    typename MatType,
+    typename VecType
+>
+class ConnectionTraits<
+    FullselfConnection<InputLayerType, OutputLayerType, OptimizerType,
+    WeightInitRule, MatType, VecType> >
+{
+ public:
+  static const bool IsSelfConnection = false;
+  static const bool IsFullselfConnection = true;
+};
+
+}; // namespace ann
+}; // namespace mlpack
+
+#endif