Tensorboard with AzureML

launch_tensorboard_azureml.py

'''
https://docs.microsoft.com/en-us/azure/machine-learning/how-to-monitor-tensorboard

make sure to install
pip install tensorflow azureml-tensorboard

Tensorboard logs need to be written in the logs folder by AML training runs.
SummaryWriter(log_dir="logs/test_run")
'''

from azureml.tensorboard import Tensorboard
from azureml.core import Workspace, Experiment
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('experiment')
parser.add_argument('--workspace-config', default="config.json")

args = parser.parse_args()
print(args)


ws = Workspace.from_config(path=args.workspace_config)
print('=' * 40)
print(ws)

exp = Experiment(ws, args.experiment)
print(exp)
runs = []
for run in exp.get_runs():
    if run.get_status() == 'Failed':
        continue
    print(run.id, run.get_status())
    runs.append(run)

local_root = 'tb_local_root'
tb = Tensorboard(runs, local_root=local_root)
print(f"Tensorboard logs will be downloaded to {local_root}")

tb.start()

# Make sure to keep the program running (or use it on a Jupyter notebook)
while True:
    input_data = input("Stop the tensorboard server? [y]")
    if input_data in ['y']:
        break

tb.stop()