tensorboard with custom docker image without notebook

Question

Hi,  
Is it possible to use tensorboard with a custom docker image without using a notebook ? Is there any other method to monitor the training process ? I'm using the tensorflow object detection API and currently exposing metrics (only loss) from cloudwatch using a regex but I'd like a more detailed way like tensorboar.. is that possible ??  
Thanks

Answer

Yes, it is possible to use tensorboard outside of the SageMaker notebooks.  
Here is an example https://github.com/awslabs/amazon-sagemaker-examples/blob/master/sagemaker-python-sdk/keras_script_mode_pipe_mode_horovod/tensorflow_keras_CIFAR10.ipynb that uses TensorBoard to compare the training jobs.  
All of it can be run locally.

Answer

For anyone interested I made a tutorial for this:  
  
https://github.com/roccopietrini/TFSagemakerDetection  
  
Edited by: rokk07 on Jan 25, 2021 2:28 AM

tensorboard with custom docker image without notebook

Contenido relevante