azureml history logging to output dir

Asked Oct 10 '19 at 10:34

Active Oct 10 '19 at 19:34

Viewed 248 times

When running my Pytorch Estimator in the cloud, I get this output regarding the recommended way to save my training scripts output to a folder called './outputs'. This folder is created by my script, and is in the correct root-folder provided. It has output.

However the estimator returns this, while preparing the container: Starting the daemon thread to refresh tokens in background for process with pid = 124 Warning: Unable to import azureml.history. Output collection disabled.

all I provide my estimator is my training script, conda env.yml and a compute target.

It completes the training script successfully, not errors.

Any ideas why I get this ?

edited Oct 10 '19 at 19:34

asked Oct 10 '19 at 10:34

ryan_jeff

I'm pretty sure the outputs have to be in the './outputs' folder and not the './output' folder. That might be why... – omartin2010 Oct 10 '19 at 15:46
Sorry, response was cut off... but I did use './outputs' , my post is just missing a letter – ryan_jeff Oct 10 '19 at 19:34
could you share your conda env.yml file? – Daniel Schneider Oct 11 '19 at 05:44
Please follow the document https://learn.microsoft.com/en-us/azure/machine-learning/service/how-to-train-pytorch#create-a-pytorch-estimator for creating a pytorch estimator. – Ram Oct 11 '19 at 11:13
Also, Can you please share your notebook/code that you are trying. – Ram Oct 11 '19 at 11:23

azureml history logging to output dir

0 Answers0