Closed mattdornfeld closed 6 years ago
Hi, first, Sacred can detect when your Tensorflow code inside an annotated function creates logs. The functionality is unfortunately completely independent of Sacred metrics Both Sacred extensions were written by me, but at the time of writing them, the Tensorflow API was changing with almost every release so I didn't even attempt to make them work together. So no, sacred metrics are not converted to tensorflow logs (nor vice versa).
To your next question: The latest development version of Sacredboard works with Tensorboard again.
To install it, you have to follow instructions for getting the development version here.
Sacredboard expects to find the tensorboard
command on the system PATH. You can start Tensorboard by choosing a captured log directory from the experiment detail view.
I've seen some references to being able to do this, but I've yet to find a full example of how the two processes work together. What's the state of this feature? Is it working? Does Sacredboard automatically start Tensorboard? How do I make sure that metrics recorded by Sacred are then visualized by Tensorboard?