Hints

Good Granite Questions

Details

You can share your hyperparameters and results with the model and ask for suggestion for improvements :)

Hint #1

Click to reveal Hint #1

Each time you run the cells, your previous experiment results will disappear like baklava crumbs on a plate. If you want to revisit your results to compare, make sure to save your results to Model Registry before starting a new experiment!

Hint #2

Click to reveal Hint #1

Does your workbench know where your bucket is? Make sure they’re properly connected.

Hint #3 (Complete solution)

Click to reveal Hint #3

Each time you complete the Notebook, you’ve performed an experiment. To save your experiments in the Model Registry:

  1. Open the OpenShift AI Dashboard.

  2. Navigate to Model Registry from the menu on the right-hand side.

  3. Click Register Model and fill in the mandatory fields.

To store hyperparameters, follow these steps:

  1. Open the version you just registered.

  2. Click Add Property to add key-value pairs.

  3. You can use the following format for storing hyperparameters:

Epochs: 100
Batch Size: 16
Learning Rate: 0.001
Hidden Layer Units: 64
Mean Squared Error: 60133219.1447
Mean Absolute Error: 6949.6542
Mean Squared Error (MSE): 60133219.14469852
Mean Absolute Error (MAE): 6949.654246425216

+

model registry

Alternatively, you can use the Python library of Model Registry if you prefer 😉

After you register three different version of the same model, you can deploy the best performing one.

To deploy use the following options:

  • Model deployment name: baklava

  • Serving runtime: OpenVino Model Server

  • Model framework: onnx - 1

  • Model location:

    > Select from existing data connection and pick `models`
> Path: `models/<your-version>/baklava`

Then, if you prefer, you can use 2-baklava_request_model.ipynb notebook to see how many kg of baklava a give region is expected to have!