Closed dinhanhx closed 3 years ago
Hi,
First you will need to enable tensorboard by training.tensorboard=True
. This will enable tensorboard logging into <save_dir>/logs
folder. Now, you can run tensorboard there with tensorboard --port 5901 --logdir .
.
Hi, First you will need to enable tensorboard by
training.tensorboard=True
. This will enable tensorboard logging into<save_dir>/logs
folder. Now, you can run tensorboard there withtensorboard --port 5901 --logdir .
.
So this means if I forget to enable before training, I can't visualize with tensorboard?
In my opinion, there should be a way to convert train.log
file to something that Tensorboard can read.
For the moment, yeah it is not possible as the events for tensorboard are generate at runtime. You can indeed log them back in or use some other tool to plot the curves by reading train.log. Here is some sample code to read and write a file to plotly: https://gist.github.com/apsdehal/1d348bb5c1a179c430f9676d4c2972f0
For the moment, yeah it is not possible as the events for tensorboard are generate at runtime. You can indeed log them back in or use some other tool to plot the curves by reading train.log. Here is some sample code to read and write a file to plotly: https://gist.github.com/apsdehal/1d348bb5c1a179c430f9676d4c2972f0
Thank you, it's certainly useful, I will use it.
Closing this as resolved with the script provided.
❓ Questions and Help
How can I use tensorboard to visualize
train.log
insave
folder produced by this commandmmf_run config=projects/hateful_memes/configs/mmbt/defaults.yaml model=mmbt dataset=hateful_memes
?