Tensorboard: Specify which scopes are removed from TensorBoard main graph by default
Migrated from: https://github.com/tensorflow/tensorflow/issues/10421
The heuristics used by TensorBoard to determine which scopes are added to the "main graph" and which are not by default often results in scopes being included that obscure the underlying network structure.
It would be nice to have an additional named parameter for
tf.name_scopethat indicated whether that scope should be removed (or included) in the main graph by default.
decentralion
All 8 comments
I would love to see this as well.
Duplicate of this https://github.com/tensorflow/tensorboard/issues/74 ?
sleighsoft
on 15 Jan 2018
+1 Would love this feature.
honkaboy
on 1 Feb 2018
+1 I spend a lot of time reconfiguring the graphs and adding scopes back to the main graph.
jcomfort4
on 13 Feb 2018
+1 to set name/variable scopes to go to either main or auxiliary graph. Or at least save the settings and layout across TensorBoard sessions.
mathemage
on 19 Apr 2018
+1 Any progress?
suisuiwudi
on 6 Jul 2018
No progress to report I'm afraid, though I agree this would be useful.
nfelt
on 7 Jul 2018
is this something where contributions would be welcomed or there are some architectural considerations that make it difficult?
jleni
on 4 Aug 2018
Probably just some kind of extension to tf.name_scope. E.g. tf.name_scope(main_graph=True) that will be inherited to sub scopes and then 1 extra field for the respective protobuf file + some handling logic within tensorboard
sleighsoft
on 5 Aug 2018
Related issues
iron9light
路
4Comments
yfarjoun
路
3Comments
datlife
路
4Comments
bestazad
路
3Comments
OverLordGoldDragon
路
3Comments
Most helpful comment
No progress to report I'm afraid, though I agree this would be useful.