C++生成别的类对象，结果发现虽然成功#include了那个类，但编译报错找不到这个类

[zhengkuxu]

#ifndef RANDOMDQN_HPP
#define RANDOMDQN_HPP

#include <mlpack/prereqs.hpp>

#include <mlpack/core.hpp>

//#include <mlpack/methods/ann/ffn.hpp>
#include <mlpack/methods/ann/init_rules/gaussian_init.hpp>
#include <mlpack/methods/ann/layer/layer.hpp>
//#include <mlpack/methods/reinforcement_learning/q_learning.hpp>

//#include "veins/base/RLmethod/ffn.hpp"
//#include "veins/base/RLmethod/ann/init_rules/gaussian_init.hpp"
//#include "veins/base/RLmethod/ann/layer/layer.hpp"
#include "veins/base/RLmethod/myffn.hpp"
#include "veins/base/RLmethod/my_q_learning.hpp"
#include <mlpack/methods/reinforcement_learning/environment/mountain_car.hpp>
#include <mlpack/methods/reinforcement_learning/environment/cart_pole.hpp>
#include <mlpack/methods/reinforcement_learning/policy/greedy_policy.hpp>
#include <mlpack/core/optimizers/adam/adam_update.hpp>
#include <mlpack/core/optimizers/rmsprop/rmsprop_update.hpp>
#include <mlpack/methods/reinforcement_learning/training_config.hpp>

using namespace mlpack;
using namespace mlpack::ann;
using namespace mlpack::optimization;
using namespace mlpack::rl;

class randomdqn {

public:
//    FFN tempmodel = new FFN(MeanSquaredError<>(), GaussianInitialization(0, 0.001));
//    GreedyPolicy temppolicy = new GreedyPolicy(1.0,1000,0.1);

    MYQLearning<CartPole, MYFFN, RMSPropUpdate, GreedyPolicy> agent;

public:
    randomdqn() {
        //! Test Double DQN in Cart Pole task.
        // Set up the network.

        MYFFN<MeanSquaredError<>, GaussianInitialization> model(MeanSquaredError<>(), GaussianInitialization(0, 0.001));
        model.Add<Linear<>>(4, 20);
        model.Add<ReLULayer<>>();
        model.Add<Linear<>>(20, 20);
        model.Add<ReLULayer<>>();
        model.Add<Linear<>>(20, 2);

        // Set up the policy and replay method.
        GreedyPolicy<CartPole> policy(1.0,1000,0.1);
        RandomReplay<CartPole> replayMethod(10, 10000);

        TrainingConfig config;
        config.StepSize() = 0.01;
        config.Discount() = 0.9;
        config.TargetNetworkSyncInterval() = 100;
        config.ExplorationSteps() = 100;
        config.DoubleQLearning() = false;
        config.StepLimit() = 200;

        // Set up the DQN agent.

//        MYQLearning<CartPole, decltype(model), RMSPropUpdate, decltype(policy)> tempagent(
//                std::move(config), std::move(model), std::move(policy),
//                std::move(replayMethod));
//
//        agent = tempagent;

        agent = new MYQLearning<CartPole, decltype(model), RMSPropUpdate, decltype(policy)>(std::move(config), std::move(model), std::move(policy),std::move(replayMethod));

    };

    ~randomdqn() {

    };
};

#endif

出问题的就是MYFFN这个类，虽然#include "veins/base/RLmethod/myffn.hpp"，但是MYFFN这个类实际上并没有被引入进来，不知道为什么会有这种情况？
MYFFN的代码如下（名为myffn.hpp)：

/**
 * Definition of the MYFFN class, which implements feed forward neural networks.
 */
#ifndef MLPACK_METHODS_ANN_MYFFN_HPP
#define MLPACK_METHODS_ANN_MYFFN_HPP

#include <mlpack/prereqs.hpp>

#include <mlpack/methods/ann/visitor/delete_visitor.hpp>
#include <mlpack/methods/ann/visitor/delta_visitor.hpp>
#include <mlpack/methods/ann/visitor/output_height_visitor.hpp>
#include <mlpack/methods/ann/visitor/output_parameter_visitor.hpp>
#include <mlpack/methods/ann/visitor/output_width_visitor.hpp>
#include <mlpack/methods/ann/visitor/reset_visitor.hpp>
#include <mlpack/methods/ann/visitor/weight_size_visitor.hpp>
#include <mlpack/methods/ann/visitor/copy_visitor.hpp>

#include <mlpack/methods/ann/init_rules/network_init.hpp>

#include <mlpack/methods/ann/layer/layer_types.hpp>
#include <mlpack/methods/ann/init_rules/random_init.hpp>
#include <mlpack/core/optimizers/rmsprop/rmsprop.hpp>

namespace mlpack {
namespace ann /** Artificial Neural Network. */ {

/**
 * Implementation of a standard feed forward network.
 *
 * @tparam OutputLayerType The output layer type used to evaluate the network.
 * @tparam InitializationRuleType Rule used to initialize the weight matrix.
 * @tparam CustomLayers Any set of custom layers that could be a part of the
 *         feed forward network.
 */
template<
  typename OutputLayerType = NegativeLogLikelihood<>,
  typename InitializationRuleType = RandomInitialization,
  typename... CustomLayers
>
class MYFFN
{
 public:
  //! Convenience typedef for the internal model construction.
  using NetworkType = MYFFN<OutputLayerType, InitializationRuleType>;

  /**
   * Create the MYFFN object with the given predictors and responses set (this is
   * the set that is used to train the network).
   * Optionally, specify which initialize rule and performance function should
   * be used.
   *
   * If you want to pass in a parameter and discard the original parameter
   * object, be sure to use std::move to avoid unnecessary copy.
   *
   * @param outputLayer Output layer used to evaluate the network.
   * @param initializeRule Optional instantiated InitializationRule object
   *        for initializing the network parameter.
   */
  MYFFN(OutputLayerType outputLayer = OutputLayerType(),
      InitializationRuleType initializeRule = InitializationRuleType());

  //! Copy constructor.
  MYFFN(const MYFFN&);

  //! Move constructor.
  MYFFN(MYFFN&&);

  //! Copy/move assignment operator.
  MYFFN& operator = (MYFFN);

  MYFFN(arma::mat predictors,
      arma::mat responses,
      OutputLayerType outputLayer = OutputLayerType(),
      InitializationRuleType initializeRule = InitializationRuleType());

  //! Destructor to release allocated memory.
  ~MYFFN();

  template<typename OptimizerType>
  void Train(arma::mat predictors,
             arma::mat responses,
             OptimizerType& optimizer);

  template<typename OptimizerType = mlpack::optimization::RMSProp>
  void Train(arma::mat predictors, arma::mat responses);

  void Predict(arma::mat predictors, arma::mat& results);

  double Evaluate(const arma::mat& parameters);

  double Evaluate(const arma::mat& parameters,
                  const size_t begin,
                  const size_t batchSize,
                  const bool deterministic);

  double Evaluate(const arma::mat& parameters,
                  const size_t begin,
                  const size_t batchSize)
  {
    return Evaluate(parameters, begin, batchSize, true);
  }

  void Gradient(const arma::mat& parameters,
                const size_t begin,
                arma::mat& gradient,
                const size_t batchSize);

  void Shuffle();

  template <class LayerType, class... Args>
  void Add(Args... args) { network.push_back(new LayerType(args...)); }

  void Add(LayerTypes<CustomLayers...> layer) { network.push_back(layer); }

  //! Return the number of separable functions (the number of predictor points).
  size_t NumFunctions() const { return numFunctions; }

  //! Return the initial point for the optimization.
  const arma::mat& Parameters() const { return parameter; }
  //! Modify the initial point for the optimization.
  arma::mat& Parameters() { return parameter; }

  /**
   * Reset the module infomration (weights/parameters).
   */
  void ResetParameters();

  //! Serialize the model.
  template<typename Archive>
  void serialize(Archive& ar, const unsigned int /* version */);

  void Forward(arma::mat inputs, arma::mat& results);

  double Backward(arma::mat targets, arma::mat& gradients);

 private:

  void Forward(arma::mat&& input);

  void ResetData(arma::mat predictors, arma::mat responses);

  void Backward();

  void Gradient(arma::mat&& input);

  void ResetDeterministic();

  /**
   * Reset the gradient for all modules that implement the Gradient function.
   */
  void ResetGradients(arma::mat& gradient);

  /**
   * Swap the content of this network with given network.
   *
   * @param network Desired source network.
   */
  void Swap(MYFFN& network);

  //! Instantiated outputlayer used to evaluate the network.
  OutputLayerType outputLayer;

  //! Instantiated InitializationRule object for initializing the network
  //! parameter.
  InitializationRuleType initializeRule;

  //! The input width.
  size_t width;

  //! The input height.
  size_t height;

  //! Indicator if we already trained the model.
  bool reset;

  //! Locally-stored model modules.
  std::vector<LayerTypes<CustomLayers...> > network;

  //! The matrix of data points (predictors).
  arma::mat predictors;

  //! The matrix of responses to the input data points.
  arma::mat responses;

  //! Matrix of (trained) parameters.
  arma::mat parameter;

  //! The number of separable functions (the number of predictor points).
  size_t numFunctions;

  //! The current error for the backward pass.
  arma::mat error;

  //! THe current input of the forward/backward pass.
  arma::mat currentInput;

  //! Locally-stored delta visitor.
  DeltaVisitor deltaVisitor;

  //! Locally-stored output parameter visitor.
  OutputParameterVisitor outputParameterVisitor;

  //! Locally-stored weight size visitor.
  WeightSizeVisitor weightSizeVisitor;

  //! Locally-stored output width visitor.
  OutputWidthVisitor outputWidthVisitor;

  //! Locally-stored output height visitor.
  OutputHeightVisitor outputHeightVisitor;

  //! Locally-stored reset visitor.
  ResetVisitor resetVisitor;

  //! Locally-stored delete visitor.
  DeleteVisitor deleteVisitor;

  //! The current evaluation mode (training or testing).
  bool deterministic;

  //! Locally-stored delta object.
  arma::mat delta;

  //! Locally-stored input parameter object.
  arma::mat inputParameter;

  //! Locally-stored output parameter object.
  arma::mat outputParameter;

  //! Locally-stored gradient parameter.
  arma::mat gradient;

  //! Locally-stored copy visitor
  CopyVisitor<CustomLayers...> copyVisitor;
}; // class MYFFN

} // namespace ann
} // namespace mlpack

// Include implementation.
#include "myffn_impl.hpp"

#endif

[zhengkuxu]

另外下面这是my_q_learning.hpp的代码

/**
 * @file q_learning.hpp
 * @author Shangtong Zhang
 *
 * This file is the definition of MYQLearning class,
 * which implements Q-Learning algorithms.
 *
 * mlpack is free software; you may redistribute it and/or modify it under the
 * terms of the 3-clause BSD license.  You should have received a copy of the
 * 3-clause BSD license along with mlpack.  If not, see
 * http://www.opensource.org/licenses/BSD-3-Clause for more information.
 */
#ifndef MLPACK_METHODS_RL_MY_Q_LEARNING_HPP
#define MLPACK_METHODS_RL_MY_Q_LEARNING_HPP

#include <mlpack/prereqs.hpp>

//#include "replay/random_replay.hpp"
//#include "training_config.hpp"
#include <mlpack/methods/reinforcement_learning/replay/random_replay.hpp>
#include <mlpack/methods/reinforcement_learning/training_config.hpp>

namespace mlpack {
namespace rl {

/**
 * Implementation of various Q-Learning algorithms, such as DQN, double DQN.
 *
 * For more details, see the following:
 * @code
 * @article{Mnih2013,
 *  author    = {Volodymyr Mnih and
 *               Koray Kavukcuoglu and
 *               David Silver and
 *               Alex Graves and
 *               Ioannis Antonoglou and
 *               Daan Wierstra and
 *               Martin A. Riedmiller},
 *  title     = {Playing Atari with Deep Reinforcement Learning},
 *  journal   = {CoRR},
 *  year      = {2013},
 *  url       = {http://arxiv.org/abs/1312.5602}
 * }
 * @endcode
 *
 * @tparam EnvironmentType The environment of the reinforcement learning task.
 * @tparam NetworkType The network to compute action value.
 * @tparam UpdaterType How to apply gradients when training.
 * @tparam PolicyType Behavior policy of the agent.
 * @tparam ReplayType Experience replay method.
 */
template <
  typename EnvironmentType,
  typename NetworkType,
  typename UpdaterType,
  typename PolicyType,
  typename ReplayType = RandomReplay<EnvironmentType> >
class MYQLearning
{
 public:
  //! Convenient typedef for state.
  using StateType = typename EnvironmentType::State;

  //! Convenient typedef for action.
  using ActionType = typename EnvironmentType::Action;

  /**
   * Create the MYQLearning object with given settings.
   *
   * If you want to pass in a parameter and discard the original parameter
   * object, be sure to use std::move to avoid unnecessary copy.
   *
   * @param config Hyper-parameters for training.
   * @param network The network to compute action value.
   * @param policy Behavior policy of the agent.
   * @param replayMethod Experience replay method.
   * @param updater How to apply gradients when training.
   * @param environment Reinforcement learning task.
   */
  MYQLearning(TrainingConfig config,
            NetworkType network,
            PolicyType policy,
            ReplayType replayMethod,
            UpdaterType updater = UpdaterType(),
            EnvironmentType environment = EnvironmentType());

  /**
   * Execute a step in an episode.
   * @return Reward for the step.
   */
  double Step();

  /**
   * Execute an episode.
   * @return Return of the episode.
   */
  double Episode();

  /**
   * @return Total steps from beginning.
   */
  const size_t& TotalSteps() const { return totalSteps; }

  //! Modify the training mode / test mode indicator.
  bool& Deterministic() { return deterministic; }
  //! Get the indicator of training mode / test mode.
  const bool& Deterministic() const { return deterministic; }

 private:
  /**
   * Select the best action based on given action value.
   * @param actionValues Action values.
   * @return Selected actions.
   */
  arma::Col<size_t> BestAction(const arma::mat& actionValues);

  //! Locally-stored hyper-parameters.
  TrainingConfig config;

  //! Locally-stored learning network.
  NetworkType learningNetwork;

  //! Locally-stored target network.
  NetworkType targetNetwork;

  //! Locally-stored updater.
  UpdaterType updater;

  //! Locally-stored behavior policy.
  PolicyType policy;

  //! Locally-stored experience method.
  ReplayType replayMethod;

  //! Locally-stored reinforcement learning task.
  EnvironmentType environment;

  //! Total steps from the beginning of the task.
  size_t totalSteps;

  //! Locally-stored current state of the agent.
  StateType state;

  //! Locally-stored flag indicating training mode or test mode.
  bool deterministic;
};

} // namespace rl
} // namespace mlpack

// Include implementation
#include <veins/base/RLmethod/my_q_learning_impl.hpp>
#endif

[zhengkuxu]

引用 12 楼 Saleayas 的回复:

你的 MYFN 是一个模板，而你用的时候，却需要一个类。 使用 MYFN<...> 代替 MYFN

不行呀，直接在OMNET++的界面就提示语法错误了。。。。

[zhengkuxu]

自顶！！！ 现在最关键的就是怎么在randomdqn的C++头文件里表示出MYFFN这个坑爹的类的类型（加了Add之后的），有谁知道怎么实现呀？这个randomdqn里的agent是我要分布式建立对象的，所以必须得是成员变量才行。。。 论坛大神帮帮忙呀！！！

[Saleayas]

你的 MYFN 是一个模板，而你用的时候，却需要一个类。 使用 MYFN<...> 代替 MYFN

[Saleayas]

MYQLearning<CartPole, MYFFN<MeanSquaredError<>, GaussianInitialization>, RMSPropUpdate, decltype(policy)> agent

[Saleayas]

MYQLearning< CartPole, MYFFN< MeanSquaredError< >, GaussianInitialization >, RMSPropUpdate, GreedyPolicy<CartPole> > agent;

[zhengkuxu]

引用 20 楼 Saleayas 的回复:

你的构造呢？ 样例里面的构造是带参数的，你有吗？

哪个的构造？

[Saleayas]

randomdqn::randomdqn() : agent(std::move(config), std::move(model), std::move(policy),std::move(replayMethod)) // 类似这样的构造。 // 看看 agent 的构造函数。 {}

[zhengkuxu]

引用 22 楼 Saleayas 的回复:

randomdqn::randomdqn() : agent(std::move(config), std::move(model), std::move(policy),std::move(replayMethod)) // 类似这样的构造。 // 看看 agent 的构造函数。 {}

可是这个有什么用呢？这不是把问题又向调用randomdqn的那一层传递了吗？

[Saleayas]

那你就像一楼的程序那样使用 new 来创建，这样使用指针。 MYQLearning< CartPole, MYFFN< MeanSquaredError< >, GaussianInitialization >, RMSPropUpdate, GreedyPolicy<CartPole> > *agent; 在析构函数里面添加 delete 就可以了。

[zhengkuxu]

引用 16 楼 Saleayas 的回复:

MYQLearning< CartPole, MYFFN< MeanSquaredError< >, GaussianInitialization >, RMSPropUpdate, GreedyPolicy<CartPole> > agent;

按你的代码报错如下：

[zhengkuxu]

引用 16 楼 Saleayas 的回复:

MYQLearning< CartPole, MYFFN< MeanSquaredError< >, GaussianInitialization >, RMSPropUpdate, GreedyPolicy<CartPole> > agent;

我感觉只要在 MYFFN的泛型列表里表示出CustomLayers的类型就好了，但我实在不知道该怎么表示。。。 大神教教我！

[zhengkuxu]

引用 18 楼 xiaoxiao123jun 的回复:

[quote=引用 16 楼 Saleayas 的回复:] MYQLearning< CartPole, MYFFN< MeanSquaredError< >, GaussianInitialization >, RMSPropUpdate, GreedyPolicy<CartPole> > agent;

我感觉只要在 MYFFN的泛型列表里表示出CustomLayers的类型就好了，但我实在不知道该怎么表示。。。 大神教教我！[/quote] 我感觉搞这个mlpack下的reinforcement_learning项目的人写的这个test程序真是个大坑啊，他直接用的decltype表示了这个神奇的MYFFN的类型。。。 C++虽然可以让你在头文件里建一个MYFFN，但是不可能实现.Add操作呀。。。。 烦死我了

[Saleayas]

你的构造呢？ 样例里面的构造是带参数的，你有吗？

[zhengkuxu]

引用 24 楼 Saleayas 的回复:

那你就像一楼的程序那样使用 new 来创建，这样使用指针。 MYQLearning< CartPole, MYFFN< MeanSquaredError< >, GaussianInitialization >, RMSPropUpdate, GreedyPolicy<CartPole> > *agent; 在析构函数里面添加 delete 就可以了。

太神奇了，按照你的办法真的就不报错了 大神收下我的膝盖，分数全部给你！ 大神能不能告诉我为什么这样就行了呢？

[百合杰]

引用 25 楼 zhao4zhong1 的回复:

模板是语法糖。 语法糖越甜，编译调试查错越苦！ 把有限的生命浪费在品尝/品鉴无穷多种的语法糖中，我认为不值当。 模板技术比不上代码生成技术。我觉得。

啥叫代码生成技术 举个栗子？

[赵4老师]

引用 27 楼 kobehahaha 的回复:

[quote=引用 25 楼 zhao4zhong1 的回复:] 模板是语法糖。 语法糖越甜，编译调试查错越苦！ 把有限的生命浪费在品尝/品鉴无穷多种的语法糖中，我认为不值当。 模板技术比不上代码生成技术。我觉得。

啥叫代码生成技术 举个栗子？[/quote] 比如google protobuf

[赵4老师]

模板是语法糖。 语法糖越甜，编译调试查错越苦！ 把有限的生命浪费在品尝/品鉴无穷多种的语法糖中，我认为不值当。 模板技术比不上代码生成技术。我觉得。

[鄢老]

类MYFFN写的有问题,且不说里面的模板没有独立出来 既然是hpp文件,那么类函数有声明必须得有实现,没见到你的实现,比如析构函数~MYFFN() 建议楼主,这种类比较大的,就分为.h和.cpp吧

[zhengkuxu]

已结贴，但是不知道为啥还能继续回复