Model file is not found for Registration of model in training Pipeline.

Question

question

Mike-Ubezzi asked May 13, '20 azure-cxp-api edited May 26, '20

Model file is not found for Registration of model in training Pipeline.

"We want the model to automatically register model every time there is a new model. we created the model in the process and write it out to a pipeline data set.To persist it then we upload and read it for registration.

We are using ./output to send the file to output. The issue is that it cannot find it in the file path . How can we validate its existence? "

[Note: As we migrate from MSDN, this question has been posted by an Azure Cloud Engineer as a frequently asked question] Source: MSDN

azure-machine-learning

5 |1600 characters needed characters left characters exceeded

Attachments: Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

Answer 1 · 2020-05-13T09:04:19.693Z

ramr-msft answered May 13, '20 ramr-msft edited May 13, '20

Can you verify that the script that is actually writing the model file to the location you expect:

 with open(model_name, 'wb') as file:
        joblib.dump(value = model, filename = os.path.join('./outputs/', model_name))

Inside in your train python script, you just need to do something like this:

persist the model to the local machine

 tf.saved_model.save(model,'./outputs/model/')

register the model with run object

 run.register_model(model_name,'./outputs/model/')

Source: MSDN

5 |1600 characters needed characters left characters exceeded

Attachments: Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

question