pytorch-lightning: [Trainer] flush_logs_every_n_steps not working

🐛 Bug

HI all ! thanks a lot for this great module 😃

Trainer has this neat argument flush_logs_every_n_steps, and does indeed take it into account before calling logger.save() in training_loop.py. Yet, the LoggerConnector flushes at each log_metrics call by calling self.trainer.logger.save().

Is that the expected behaviour of flush_logs_every_n_steps? Shouldn’t it be something like

if self.trainer.logger is not None:
    if self.trainer.is_global_zero:
        self.trainer.logger.agg_and_log_metrics(scalar_metrics, step=step)
        if self.should_flush_logs:
            self.trainer.logger.save()

in logger_connector.py ?

To Reproduce

Print when Logger.save is called.

Expected behavior

Only call logger.save() every n steps.

Environment

I’ve tried to reproduce the bug in Colab but I get an error as test_tube is not installed even though I did pip install test_tube .

• CUDA:
  • GPU:
  • available: False
  • version: 10.2
• Packages:
  • numpy: 1.19.4
  • pyTorch_debug: True
  • pyTorch_version: 1.7.0
  • pytorch-lightning: 1.0.6
  • tqdm: 4.51.0
• System:
  • OS: Linux
  • architecture:
    • 64bit
  • processor: x86_64
  • python: 3.7.9
  • version: #1 SMP Tue Aug 11 16:36:14 UTC 2020

Cheers, Emile