Tensorboard: Tensorboard.dev upload to specified experiment ID

Created on 25 Aug 2020  路  2Comments  路  Source: tensorflow/tensorboard

Currently, there is no way to upload new data to an existing experiment once the tensorboard client has been interrupted. However, this use case comes up very often (e.g. running a bunch of hyperparameter search experiments, then another bunch, or implementing a new model a few days later and wanting to compare). The only solution is to delete the older experiment and start a new experiment, assuming that the data exist on the same machine. This brings us to the second benefit: in case one is performing experiments on different machines, it is a lot more convenient to upload all to the same URL/experiment ID, than to aggregate the data (or mount the remote drives) on a single machine and then upload them to an experiment.
The proposed functionality would be evoked like this:
tensorboard dev update --experiment_id EXPERIMENT_ID --logdir logs

tbdev feature

Most helpful comment

Thank you for your feature request @gzerveas, this is a request we have heard from several users, and something we will prioritize.

All 2 comments

Thank you for your feature request @gzerveas, this is a request we have heard from several users, and something we will prioritize.

up this, would be really cool to have this functionality

Was this page helpful?
0 / 5 - 0 ratings