caffe的LSTM层实现与解析（非官方版）

最近看CTPN的论文时看到了demo中实现的LSTM层，这和caffe官方的基于recurrent层的不一样，是独立的。我对它做了一些注解，在这里贴出来方便以后查看。

#include <vector>
#include "caffe/blob.hpp"
#include "caffe/common.hpp"
#include "caffe/filler.hpp"
#include "caffe/layer.hpp"
#include "caffe/util/math_functions.hpp"
#include "caffe/layers/lstm_layer.hpp"


namespace caffe {

template <typename Dtype>
inline Dtype sigmoid(Dtype x) {
  return 1. / (1. + exp(-x));
}

template <typename Dtype>
void caffe_bound(const int N, const Dtype* a, const Dtype min, 
    const Dtype max, Dtype* y) {
  for (int i = 0; i < N; ++i) {
    y[i] = std::min(std::max(a[i], min), max);
  }
}

template <typename Dtype>
void LstmLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
      const vector<Blob<Dtype>*>& top) {//input: W*(H*N)*(C*3*3)
  clipping_threshold_ = this->layer_param_.lstm_param().clipping_threshold();
  N_ = bottom[0]->shape(1); // batch_size
  H_ = this->layer_param_.lstm_param().num_output(); // number of hidden units
  I_ = bottom[0]->count(2); // input dimension

  // Check if we need to set up the weights
  if (this->blobs_.size() > 0) {
    LOG(INFO) << "Skipping parameter initialization";
  } else {
    this->blobs_.resize(3);   //three weights blobs: W_xh, W_hh, b_xh
    shared_ptr<Filler<Dtype> > weight_filler(GetFiller<Dtype>(
        this->layer_param_.lstm_param().weight_filler())); //get the appointed weight_filler 
 
    // input-to-hidden weights    W_xh
    // Intialize the weight
    vector<int> weight_shape; 
    weight_shape.push_back(4*H_);  //there are 4 gates need Wxh
    weight_shape.push_back(I_);  //{4*H_, I_}
    this->blobs_[0].reset(new Blob<Dtype>(weight_shape));
    weight_filler->Fill(this->blobs_[0].get());

    // hidden-to-hidden weights   W_hh
    // Intialize the weight
    weight_shape.clear();
    weight_shape.push_back(4*H_);  //there are four gates need Whh, too
    weight_shape.push_back(H_);  //(4*H_, H_)
    this->blobs_[1].reset(new Blob<Dtype>(weight_shape));
    weight_filler->Fill(this->blobs_[1].get());

    // If necessary, intiialize and fill the bias term   b_xh
    vector<int> bias_shape(1, 4*H_);  //{4*H_}
    this->blobs_[2].reset(new Blob<Dtype>(bias_shape));
    shared_ptr<Filler<Dtype> > bias_filler(GetFiller<Dtype>(
        this->layer_param_.lstm_param().bias_filler()));
    bias_filler->Fill(this->blobs_[2].get());
  }  // parameter initialization
  this->param_propagate_down_.resize(this->blobs_.size(), true);

  
}

template <typename Dtype>
void LstmLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
      const vector<Blob<Dtype>*>& top) {//input: W*(H*N)*(C*3*3)
  // Figure out the dimensions
  T_ = bottom[0]->shape(0); // length of sequence
  N_=bottom[0]->shape(1);  //batch_size
  CHECK_EQ(bottom[0]->count(2), I_) << "Input size "
    "incompatible with inner product parameters.";
  vector<int> original_top_shape;
  original_top_shape.push_back(T_);
  original_top_shape.push_back(N_);
  original_top_shape.push_back(H_);
  top[0]->Reshape(original_top_shape); // output shape:{T_, N_, H_}

  vector<int> cell_shape; // memory cell shape: {N_, H_}
  cell_shape.push_back(N_);
  cell_shape.push_back(H_);
  c_0_.Reshape(cell_shape);   //cell state at time 0
  h_0_.Reshape(cell_shape);   //hidden layer output at time 0
  c_T_.Reshape(cell_shape);   //cell state at time T
  h_T_.Reshape(cell_shape);   //hidden layer output at tiem T
  h_to_h_.Reshape(cell_shape);

  vector<int> gate_shape;
  gate_shape.push_back(N_);
  gate_shape.push_back(4);
  gate_shape.push_back(H_);
  h_to_gate_.Reshape(gate_shape);  //{N_, 4, H_}

  // Gate initialization
  gate_shape.clear();
  gate_shape.push_back(T_);
  gate_shape.push_back(N_);
  gate_shape.push_back(4);
  gate_shape.push_back(H_);
  pre_gate_.Reshape(gate_shape);  // the input of gates
  gate_.Reshape(gate_shape);   //  the outout of gates (namely after nonlinearity): shape{T_, N_, 4, H_}
  
  vector<int> top_shape;
  top_shape.push_back(T_);   // {T_, N_, H_}
  top_shape.push_back(N_);
  top_shape.push_back(H_);
  cell_.Reshape(top_shape);
  top_.Reshape(top_shape);
  top_.ShareData(*top[0]);
  top_.ShareDiff(*top[0]);
  
  // Set up the bias multiplier
  vector<int> multiplier_shape(1, N_*T_);
  bias_multiplier_.Reshape(multiplier_shape);
  caffe_set(bias_multiplier_.count(), Dtype(1), 
    bias_multiplier_.mutable_cpu_data());
}

template <typename Dtype>
void LstmLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
    const vector<Blob<Dtype>*>& top) {
  CHECK_EQ(top[0]->cpu_data(), top_.cpu_data());
  Dtype* top_data = top_.mutable_cpu_data();
  const Dtype* bottom_data = bottom[0]->cpu_data();
  const Dtype* clip = NULL;
  if (bottom.size() > 1) {
    clip = bottom[1]->cpu_data();    //cont blob, shape:{T, N}  default is set to be {0,0,...,0; 1,1,...,1; ...; 1,1,..,1}
    CHECK_EQ(bottom[1]->num(), bottom[1]->count());
  }
  const Dtype* weight_i = this->blobs_[0]->cpu_data();
  const Dtype* weight_h = this->blobs_[1]->cpu_data();
  const Dtype* bias = this->blobs_[2]->cpu_data();
  Dtype* pre_gate_data = pre_gate_.mutable_cpu_data();
  Dtype* gate_data = gate_.mutable_cpu_data();
  Dtype* cell_data = cell_.mutable_cpu_data();
  Dtype* h_to_gate = h_to_gate_.mutable_cpu_data();

  // Initialize previous state
  if (clip) {
    caffe_copy(c_0_.count(), c_T_.cpu_data(), c_0_.mutable_cpu_data());  //make c_T = c_0
    caffe_copy(h_0_.count(), h_T_.cpu_data(), h_0_.mutable_cpu_data());  //make h_T = h_0
  }
  else {
    caffe_set(c_0_.count(), Dtype(0.), c_0_.mutable_cpu_data());  // initialize  c_0 = 0
    caffe_set(h_0_.count(), Dtype(0.), h_0_.mutable_cpu_data());  // initialize  h_0 = 0
  }

  // Compute input to hidden forward propagation    
  //  function: C(matrix) ← αAB + βC  
  //  void caffe_cpu_gemm<float>(const CBLAS_TRANSPOSE TransA,

  //  const CBLAS_TRANSPOSE TransB, const int M, const int N, const int K,

  //  const float alpha, const float* A, const float* B, const float beta,

  //  float* C) 
  //
  caffe_cpu_gemm(CblasNoTrans, CblasTrans, T_*N_, 4*H_, I_, Dtype(1.),
      bottom_data, weight_i, Dtype(0.), pre_gate_data);    //pre_gate_data = bottom_data * weight_i
  caffe_cpu_gemm(CblasNoTrans, CblasNoTrans, T_*N_, 4*H_, 1, Dtype(1.),
      bias_multiplier_.cpu_data(), bias, Dtype(1.), pre_gate_data);  //pre_gate_data = bias_multiplier_ * bias + pre_gate_data

  // Compute recurrent forward propagation   share the same weights in RNN
  for (int t = 0; t < T_; ++t) {
    Dtype* h_t = top_data + top_.offset(t); //get the h_t point at time t
    Dtype* c_t = cell_data + cell_.offset(t); //get the c_t point at tiem t
    Dtype* pre_gate_t = pre_gate_data + pre_gate_.offset(t); // get the pre_gate_t point at time t
    Dtype* gate_t = gate_data + gate_.offset(t);  // get the gate_t point at time t
    Dtype* h_to_gate_t = h_to_gate;
    const Dtype* clip_t = clip ? clip + bottom[1]->offset(t) : NULL;
    const Dtype* h_t_1 = t > 0 ? (h_t - top_.offset(1)) : h_0_.cpu_data();
    const Dtype* c_t_1 = t > 0 ? (c_t - cell_.offset(1)) : c_0_.cpu_data();

    // Hidden-to-hidden propagation
    caffe_cpu_gemm(CblasNoTrans, CblasTrans, N_, 4*H_, H_, Dtype(1.), 
        h_t_1, weight_h, Dtype(0.), h_to_gate); // h_to_gate = h_t_1 * weight_h 

    for (int n = 0; n < N_; ++n) {
      const bool cont = clip_t ? clip_t[n] : t > 0;
      if (cont) {
        caffe_add(4*H_, pre_gate_t, h_to_gate, pre_gate_t);  //pre_gate = pre_gate_t + h_to_gate
      }
      for (int d = 0; d < H_; ++d) { // for each node in hidden layer
        // Apply nonlinearity: activation functions
        gate_t[d] = sigmoid(pre_gate_t[d]);  //the output of 'input gate' :i(t)
        gate_t[H_ + d] = cont ? sigmoid(pre_gate_t[H_ + d]) : Dtype(0.);  //the output of 'forget gate' :f(t)
        gate_t[2*H_ + d] = sigmoid(pre_gate_t[2*H_ + d]);  //the output of 'output gate' :o(t)
        gate_t[3*H_ + d] = tanh(pre_gate_t[3*H_ + d]);   // the input after tanh: g(t)

        // Compute cell : c(t) = f(t)*c(t-1) + i(t)*g(t)
        c_t[d] = gate_t[H_ + d] * c_t_1[d] + gate_t[d] * gate_t[3*H_ + d];
		// Compute hidden layer output: h(t) = o(t).*tanh[c(t)]
        h_t[d] = gate_t[2*H_ + d] * tanh(c_t[d]);
      }
      
      h_t += H_;   //move point to the next n's place
      c_t += H_;   //
      c_t_1 += H_;
      pre_gate_t += 4*H_;
      gate_t += 4*H_;
      h_to_gate_t += 4*H_;
    }
  }
  // Preserve cell state and output value for truncated BPTT
  caffe_copy(N_*H_, cell_data + cell_.offset(T_-1), c_T_.mutable_cpu_data());
  caffe_copy(N_*H_, top_data + top_.offset(T_-1), h_T_.mutable_cpu_data());
}

template <typename Dtype>
void LstmLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
    const vector<bool>& propagate_down,
    const vector<Blob<Dtype>*>& bottom) {
  const Dtype* top_data = top_.cpu_data();
  const Dtype* bottom_data = bottom[0]->cpu_data();
  const Dtype* clip = NULL;
  if (bottom.size() > 1) {
    clip = bottom[1]->cpu_data();
    CHECK_EQ(bottom[1]->num(), bottom[1]->count());
  }
  const Dtype* weight_i = this->blobs_[0]->cpu_data();
  const Dtype* weight_h = this->blobs_[1]->cpu_data();
  const Dtype* gate_data = gate_.cpu_data();
  const Dtype* cell_data = cell_.cpu_data();

  Dtype* top_diff = top_.mutable_cpu_diff();
  Dtype* pre_gate_diff = pre_gate_.mutable_cpu_diff();
  Dtype* gate_diff = gate_.mutable_cpu_diff();
  Dtype* cell_diff = cell_.mutable_cpu_diff();
  
  caffe_copy(N_*H_, c_T_.cpu_diff(), cell_diff + cell_.offset(T_-1));  

  for (int t = T_-1; t >= 0; --t) {
    Dtype* dh_t = top_diff + top_.offset(t);
    Dtype* dc_t = cell_diff + cell_.offset(t);
    Dtype* gate_diff_t = gate_diff + gate_.offset(t);
    Dtype* pre_gate_diff_t = pre_gate_diff + pre_gate_.offset(t);
    Dtype* dh_t_1 = t > 0 ? top_diff + top_.offset(t-1) : h_0_.mutable_cpu_diff();
    Dtype* dc_t_1 = t > 0 ? cell_diff + cell_.offset(t-1) : c_0_.mutable_cpu_diff();
    const Dtype* clip_t = clip ? clip + bottom[1]->offset(t) : NULL;
    const Dtype* c_t = cell_data + cell_.offset(t);
    const Dtype* c_t_1 = t > 0 ? cell_data + cell_.offset(t-1) : c_0_.cpu_data();
    const Dtype* gate_t = gate_data + gate_.offset(t);

    for (int n = 0; n < N_; ++n) {
      const bool cont = clip_t ? clip_t[n] : t > 0;
      for (int d = 0; d < H_; ++d) {
        const Dtype tanh_c = tanh(c_t[d]);
        gate_diff_t[2*H_ + d] = dh_t[d] * tanh_c;
        dc_t[d] += dh_t[d] * gate_t[2*H_ + d] * (Dtype(1.) - tanh_c * tanh_c);
        dc_t_1[d] = cont ? dc_t[d] * gate_t[H_ + d] : Dtype(0.);
        gate_diff_t[H_ + d] = cont ? dc_t[d] * c_t_1[d] : Dtype(0.);
        gate_diff_t[d] = dc_t[d] * gate_t[3*H_ + d];
        gate_diff_t[3*H_ +d] = dc_t[d] * gate_t[d];

        pre_gate_diff_t[d] = gate_diff_t[d] * gate_t[d] * (Dtype(1.) - gate_t[d]);
        pre_gate_diff_t[H_ + d] = gate_diff_t[H_ + d] * gate_t[H_ + d] 
            * (1 - gate_t[H_ + d]);
        pre_gate_diff_t[2*H_ + d] = gate_diff_t[2*H_ + d] * gate_t[2*H_ + d] 
            * (1 - gate_t[2*H_ + d]);
        pre_gate_diff_t[3*H_ + d] = gate_diff_t[3*H_ + d] * (Dtype(1.) - 
            gate_t[3*H_ + d] * gate_t[3*H_ + d]);
      }

      // Clip deriviates before nonlinearity
      if (clipping_threshold_ > Dtype(0.)) {
        caffe_bound(4*H_, pre_gate_diff_t, -clipping_threshold_, 
            clipping_threshold_, pre_gate_diff_t);
      }

      dh_t += H_;
      c_t += H_;
      c_t_1 += H_;
      dc_t += H_;
      dc_t_1 += H_;
      gate_t += 4*H_;
      gate_diff_t += 4*H_;
      pre_gate_diff_t += 4*H_;
    }
    
    // Backprop output errors to the previous time step
    caffe_cpu_gemm(CblasNoTrans, CblasNoTrans, N_, H_, 4*H_,
        Dtype(1.), pre_gate_diff + pre_gate_.offset(t), 
        weight_h, Dtype(0.), h_to_h_.mutable_cpu_data());
    for (int n = 0; n < N_; ++n) {
      const bool cont = clip_t ? clip_t[n] : t > 0;
      const Dtype* h_to_h = h_to_h_.cpu_data() + h_to_h_.offset(n);
      if (cont) {
        caffe_add(H_, dh_t_1, h_to_h, dh_t_1);
      }
    }
  }
 
  if (this->param_propagate_down_[0]) {
    // Gradient w.r.t. input-to-hidden weight
    caffe_cpu_gemm(CblasTrans, CblasNoTrans, 4*H_, I_, T_*N_, Dtype(1.),
        pre_gate_diff, bottom_data, Dtype(1.), this->blobs_[0]->mutable_cpu_diff());
  }

  if (this->param_propagate_down_[1]) {
    // Gradient w.r.t. hidden-to-hidden weight
    caffe_cpu_gemm(CblasTrans, CblasNoTrans, 4*H_, H_, (T_-1)*N_, Dtype(1.),
        pre_gate_diff + pre_gate_.offset(1), top_data, 
        Dtype(1.), this->blobs_[1]->mutable_cpu_diff());

    // Add Gradient from previous time-step
    caffe_cpu_gemm(CblasTrans, CblasNoTrans, 4*H_, H_, 1, Dtype(1.),
        pre_gate_diff, h_0_.cpu_data(), 
        Dtype(1.), this->blobs_[1]->mutable_cpu_diff());
  }
  if (this->param_propagate_down_[2]) { 
    // Gradient w.r.t. bias
    caffe_cpu_gemv(CblasTrans, T_*N_, 4*H_, Dtype(1.), pre_gate_diff,
        bias_multiplier_.cpu_data(), Dtype(1.),
        this->blobs_[2]->mutable_cpu_diff());
  }
  if (propagate_down[0]) {
    // Gradient w.r.t. bottom data
    caffe_cpu_gemm(CblasNoTrans, CblasNoTrans, T_*N_, I_, 4*H_, Dtype(1.),
        pre_gate_diff, weight_i, Dtype(0.), bottom[0]->mutable_cpu_diff());
  }
}

#ifdef CPU_ONLY
STUB_GPU(LstmLayer);
#endif

INSTANTIATE_CLASS(LstmLayer);
REGISTER_LAYER_CLASS(Lstm);

}  // namespace caffe