Class DSSMInstReader

src/model/instance_reader/dssm_inst_reader.cc:26–153

namespace embedx {

class DSSMInstReader : public EmbedInstanceReader {
 private:
  bool is_train_ = true;
  int num_neg_ = 5;
  bool add_node_ = true;

 private:
  DeepFlow flow_;
  // for training/predicting data
  vec_int_t pos_items_;
  std::vector<vec_pair_t> feats_list_;
  std::vector<vec_int_t> neg_items_list_;

  // for item feature
  vec_int_t unique_items_;
  std::vector<vec_pair_t> item_feats_list_;

  uint16_t ns_id_;
  std::unique_ptr<IndexingWrapper> indexing_wrapper_;

 public:
  DEFINE_INSTANCE_READER_LIKE(DSSMInstReader);

 protected:
  bool InitConfigKV(const std::string& k, const std::string& v) override {
    if (InstanceReaderImpl::InitConfigKV(k, v)) {
    } else if (k == "is_train") {
      auto val = std::stoi(v);
      DXCHECK(val == 1 || val == 0);
      is_train_ = val;
    } else if (k == "num_neg") {
      num_neg_ = std::stoi(v);
      DXCHECK(num_neg_ > 0);
    } else if (k == "add_node") {
      auto val = std::stoi(v);
      DXCHECK(val == 1 || val == 0);
      add_node_ = val;
    } else {
      DXERROR("Unexpected config: %s = %s.", k.c_str(), v.c_str());
      return false;
    }

    DXINFO("Instance reader argument: %s = %s.", k.c_str(), v.c_str());
    return true;
  }

  void PostInit(const std::string& /*node_config*/) override {
    ns_id_ = 0;
    indexing_wrapper_ = IndexingWrapper::Create("");
  }

 public:
  bool GetBatch(Instance* inst) override {
    return is_train_ ? GetTrainBatch(inst) : GetPredictBatch(inst);
  }

  /************************************************************************/
  /* Read batch data from file for training */