Problem Description
ubuntu16.04, tensorflow-gpu version is 2.1.0, I want to use tensorboard to record the loss update process, enter it in the terminal,
tensorboard --logdir=logs
get the URLhttp://ubuntu-acloud:xxx/
, click to enter, there is indeed a loss curve, but it has not been refreshed, only to re-enter the above command in the terminal, It is useful to get the URL and click in again, very speechless...
solution
To be clear, tensorflow2.x comes with tensorboard when it is installed. When checking tensorflow-gpu2.1 in this system, it downloads tensorboard2.1.1 when it
istensorboard --logdir=logs
installedTensorBoard 1.14.0 at http://ubuntu-acloud:xxx/ (Press CTRL+C to quit)
. , it used tensorBoard 1.14.0
and took a closer look. I entered the command in the base environmenttensorboard --logdir=logs
. In this environment, tensorBoard is 1.14.0, but the running program uses another conda environment. Switch to the running program environment. Enter againtensorboard --logdir=logs
, the loss is finally updated, and you're done!
lesson
Since multiple conda environments are configured, be sure to pay attention to environment switching when entering commands in the terminal