Search code examples
pytorchtensorboard

How to use torch.utils.tensorboard.SummaryWriter from the last interrupted events

enter image description here

As in this picture, if I want to add scalar from events.out.tfevents, but not create a new one.

How can I set the params int this code:

SummaryWriter(self, log_dir=None, comment='', purge_step=None, max_queue=10,
             flush_secs=120, filename_suffix='')
Solution

  • You should be able to run it the same way (e.g. log_dir has to be the same, tensorboard in your case).

    You have to remember to use next global step when adding scalar though.

    First run, assume it crashed at 9th step:

    from torch.utils.tensorboard import SummaryWriter

writer = SummaryWriter("my_dir")
x = range(10)
for i in x:
    writer.add_scalar("y=x", i, i)
writer.close()

    If you want to continue writing to this event file, you have to move last parameter global step by 10:

    from torch.utils.tensorboard import SummaryWriter

writer = SummaryWriter("my_dir")
x = range(10)
for i in x:
    writer.add_scalar("y=x", i, i + 10) # start from step 10
writer.close()

    Running first file, followed by the second one and opening tensorboard via tensorboard --logdir my_dir would give you:

    tensorboard_restart