PrimeHub Quickstart

Enterprise Applicable to Enterprise Edition

Community Applicable to Community Edition

1-Minute PrimeHub Demo

Brief

In this quickstart, we will learn how to use a Notebook to train/generate a simple model. Then, in the follow-up guide, we will learn how to submit a parameter-tuning job for the model tuning from same Notebook.

• Notebook is a fundamental feature of Community/Enterprise editions.
• Jobs is a feature of Enterprise edition.

Training a model on Notebook

What we need?

• The image infuseai/docker-stacks:tensorflow-notebook-v2-4-1-5b5a244c

• An instance type >= minimal requirement (CPU=1, GPU=0, Mem=2G)

• The prepared notebook file:

  Download my_model.ipynb

• Choose a group with enabled Shared Volume (a.k.a Group Volume)

Please have the image, the instance type on PrimeHub or request administrators for the assistance before we start.

Steps

1. Enter Notebook from User Portal, select the image, the instance type and start a notebook.

2. From File Browser of Notebook, navigate into the directory of <group_name> which is a Group Volume; here data-team is our working group.

3. While inside the group volume, copy/drag the downloaded my_model.ipynb there in File Browser and open it.

4. Run All Cells, it will save the model file my_model.h5 at the local.

   

So far, we have trained and generated a model file via our Notebook. This is achievable in both Community and Enterprise editions of PrimeHub.

The next stage, parameter-tuning job, is for Enterprise users. Keep your Notebook environment open, and let's continue on...

Parameter-tuning Job

Enterprise Applicable to Enterprise Edition

What we need?

• The same image as above

• The same instance type as above

• API Token

• The prepared notebook file of the example

  Download tune_params.ipynb

Steps

1. While inside the group volume, copy/drag the downloaded tune_params.ipynb and open it.

   • Click on Cell 4, from Property Inspector, we can see the notebook has set epoch as a parameter

     See Papermill site ⇗ for the detail of Parameterize.

   

2. Back to PrimeHub tab, hover over Profile icon, generate a API Token and copy it.

   

3. Back to Notebook tab, click PrimeHub dropdown on the toolbar of the opened notebook, select API Token and paste the generated token here.

   

   

4. Click PrimeHub dropdown again and select Submit Notebook as Job this time.

5. Select the instance type and the image; fill in Job Name tune_params_epoch_tf and

   

   Fill inNotebook Parameters with the code below due to epoch is parameterized.

   epoch = 2
   

6. Back to PrimeHub tab, enter Jobs and check the job tune_params_epoch_tf we submitted.

   

7. Once the job succeeded, back to Notebook, we can find a notebook tune_params-<datetime> of output by the job and the overwritten my_model.h5 after tuning.

   

We now have walked through a typical flow of training and saving a simple model file, then submitting a job of tuning the generated on Notebook. In a real project, we may want to use various combinations of jobs and notebooks to pipeline repeatable workflows.

What's Next?

As Enterprise users, we desperately want to verify the performance of our model, therefore, we can deploy models as services and query the services with testing data.

• Quick-start: PrimeHub Model Deployment

Reference

• Submit a Notebook as a Job
• Jobs Submission