Model Select and Optimize

In this directory, notebooks are provided to demonstrate how to tune and optimize hyperparameters of recommender algorithms with the utility functions (recommenders) provided in the repository.

Notebook	Description
tuning_spark_als	Step by step tutorials on how to fine tune hyperparameters for Spark based recommender model (illustrated by Spark ALS) with Spark native construct and `hyperopt` package.
azureml_hyperdrive_wide_and_deep	Quickstart tutorial on utilizing Azure Machine Learning service for hyperparameter tuning of wide-and-deep model.
azureml_hyperdrive_surprise_svd	Quickstart tutorial on utilizing Azure Machine Learning service for hyperparameter tuning of the matrix factorization method SVD from Surprise library.
nni_surprise_svd	Quickstart tutorial on utilizing the Neural Network Intelligence toolkit for hyperparameter tuning of the matrix factorization method SVD from Surprise library.
nni_ncf	Quickstart tutorial on utilizing the Neural Network Intelligence toolkit as a tool to tune the NCF model and SVD model and compare their performance against one another

Prerequisites

To run the examples running on the Azure Machine Learning service, the azureml-sdk is required. The AzureML Python SDK is already installed after setting up the conda environments from this repository (see SETUP.md).

More info about setting up an AzureML environment can be found at this link.

AzureML Workspace Configuration

AzureML workspace is the foundational block in the cloud that you use to experiment, train, and deploy machine learning models. We

set up a workspace from Azure portal and
create a config file manually.

The instructions here are based on AzureML documents about Quickstart with Azure portal and Quickstart with Python SDK where you can find more details with screenshots about the setup process.

Create a workspace

Sign in to the Azure portal by using the credentials for the Azure subscription you use.
Select Create a resource menu, search for Machine Learning service workspace select Create button.
In the ML service workspace pane, configure your workspace by entering the workspace name and resource group (or create new resource group if you don't have one already), and select Create. It can take a few moments to create the workspace.

Make a configuration file

To configure this notebook to communicate with your workspace, type in your Azure subscription id, the resource group name and workspace name to , , in the cell below. Alternatively, you can create a ./aml_config/config.json file with the following contents:

{
    "subscription_id": "<subscription-id>",
    "resource_group": "<resource-group>",
    "workspace_name": "<workspace-name>"
}

NNI Configuration

The NNI command nnictl comes installed with the conda environment. In order to use the SMAC tuner, it has to be installed first with the following command nnictl package install --name=SMAC.