You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Make it so it by default when opening Tensorboard in the default log directory runs are stored in ascending or descending order.
Pitch
The current TensorBoardLogger default version stamping does not provide a graceful view when sorting experiment runs from TensorBoard.
I think this could be easily solved by putting a default parameter for the TensorBoardLogger, something like n_version_placeholders = 4, and changing the default when getting the log_dir:
deflog_dir(self) ->str:
"""The directory for this run's tensorboard checkpoint. By default, it is named ``'version_${self.version}'`` but it can be overridden by passing a string value for the constructor's version parameter instead of ``None`` or an int. """# CHANGED else f"version_{self.version}" to f"version_{self.version:0{self.n_version_placeholders}d}"version=self.versionifisinstance(self.version, str) elsef"version_{self.version:0{self.n_version_placeholders}d}"log_dir=os.path.join(self.root_dir, version)
ifisinstance(self.sub_dir, str):
log_dir=os.path.join(log_dir, self.sub_dir)
log_dir=os.path.expandvars(log_dir)
log_dir=os.path.expanduser(log_dir)
returnlog_dir
Alternatives
timestamps...? I don't like timestamps as they are messy but would do the trick.
Description & Motivation
Make it so it by default when opening Tensorboard in the default log directory runs are stored in ascending or descending order.
Pitch
The current TensorBoardLogger default version stamping does not provide a graceful view when sorting experiment runs from TensorBoard.
I think this could be easily solved by putting a default parameter for the TensorBoardLogger, something like n_version_placeholders = 4, and changing the default when getting the log_dir:
Alternatives
timestamps...? I don't like timestamps as they are messy but would do the trick.
cc @Borda
The text was updated successfully, but these errors were encountered: