batch queue is empty (1 warnings)

[Huang-Jin]

Hello,
Thank you for your excellent work and codes. When I'm training using your code, I find that it outputs warnings every iteration, like this:

# 2018-10-24 15:19:12     3800 | total_loss:      91.93  learning_rate:    0.00025  
batch queue is empty (101 warnings)
# 2018-10-24 15:21:03     3900      0.90 ips            18:14:20 rem | total_loss:      137.2  learning_rate:    0.00025  
batch queue is empty (201 warnings)

I find that this warning comes from https://github.com/lmb-freiburg/demon/blob/devel/multivih5datareaderop/multivih5datareader.cpp.

Batch_sptr getBatch()
    {
      bool print_warning = !param.test_phase;
      Batch_sptr result = nullptr;
      while( !result )
      {
        std::lock_guard<std::mutex> lock(batches_mutex);
        if( batches.size() )
        {
          result = batches.front();
          batches.pop();
        }
        else
        {
          if(print_warning)
          {
            static int empty_count = 0;
            ++empty_count;
            if( !((empty_count-1)%100) )
              std::cerr << "batch queue is empty (" << empty_count << " warnings)\n";
          }
          print_warning = false;
        }
      }
      return result;
    }

So, does it means that there are no data input and the training is useless?
Waiting for your apply!

[benjaminum]

It means that the training has to wait for data and could be faster.