Sagemaker Sweep Monitor

sweep.py

import sagemaker
import wandb
tuner_name = "sagemaker-pytorch-190111-1056"
tuner = sagemaker.HyperparameterTuner.attach(tuner_name)
job_name = tuner.analytics().description()['TrainingJobDefinition']['StaticHyperParameters']['sagemaker_job_name']
client = tuner.sagemaker_session.sagemaker_client

api = wandb.Api()
runs = api.runs("wandb/sm-pytorch-cifar", {"config.sagemaker_job_name": job_name})
for run in runs:
  if run.summary["_step"] > 100 and run.summary["Test Acc"] < 0.5:
    client.stop_training_job(TrainingJobName=run.name.replace("algo-1", ""))