I’m currently experimenting with the new implementation of Tensorboard in Torch.Utilis and so far it works fine.
The only problem I have is that I have to restart the process every time to update the scalar graphs (maybe also the other Graph-Types, but I did not test them).
I see, that Tensorboard detect new log files, but it do not show the data until i kill and restart the process.
Is there a possability to fix this, right now?
Do the new values eventually show up after a while?
If so, you might want to experiment a bit the the --reload_interval SECONDS
argument when starting tensorboard
.
No, also after 1 hour it’s not loading new data
a minimal example is:
writer = SummaryWriter(log_dir)
for epoch i n range(epochs):
train(epoch)
loss_test = test()
writer_val.add_scalar('loss', loss_test, epoch)
writer.close()
I had the same issue. In my case I solved the issue using a previous version of tensorboard.
so, which version do you use right now?
I am using TensorBoard 1.13.1.
If i try to run it with TensorBoard in 1.13. I get the error:
ImportError: TensorBoard logging requires TensorBoard with Python summary writer installed.
This should be available in 1.14 or above.
how did you do this?
As far as I remember (I do not have access at this moment to my develop setup) I have two different venv. In one of them I have pytorch + tensorboard1.14, in the second one, I have tensorboard 1.13. I am using the second one to display my tensorboard.
I’m not sure if this work for me…
maybe I have to wait for tensorboard gets updated to 1.14. without beeing nightly build.
But thanks for your help!
anyone else an idea, before the update?
Why it is not working for you?
I think your solution is a good work arround!
But there must be a reason for the “not updating” problem
Yes, but I think it is in tensorboard side.