cford38/amiss_aml repository overview

⁠Scaling the amiss R framework with Azure Machine Learning

⁠Quest Diagnostics, Blueprint Genetics, and BlueGranite

⁠amiss R package repository: https://github.com/blueprint-genetics/amiss⁠

Since the amiss framework performs a grid search across a number of parameters, this process can be distributed across multiple nodes in a compute cluster. Using Azure Machine Learning, a cloud-based service to help build, scale, monitor, and deploy machine learning models, we can distribute this search with ease.

To learn more about Azure Machine Learning, click here⁠.

⁠How it works - HyperDrive

As part of the Azure Machine Learning Python and R SDKs, there is a hyperparameter tuning package called HyperDrive⁠.

HyperDrive allows for more efficient hyperparameter tuning by distributing combinations of parameters to separate compute contexts on a cluster. This helps to find the set of parameters that result in the best performance.

⁠Running amiss in Azure ML

1. Provision an Azure Machine Learning workspace in your Azure tenant using the following instructions: https://docs.microsoft.com/en-us/azure/machine-learning/quickstart-create-resources#create-the-workspace⁠

2. Once the Azure ML workspace is ready, create a Compute Instance as described here⁠.

3. Once the Compute Instance is up and running, click on the RStudio link to open an RStudio Server instance in your browser.

4. Upload the aml_r_sdk_hyperdrive.rmd⁠ notebook and the amiss_test_script.R⁠ script into RStudio.

5. Upload your data into RStudio under the /data directory. (You may have to create this directory first.)

6. Change the values in the the aml_r_sdk_hyperdrive.rmd notebook to fit your desired computational workload.

7. Run the cells from the aml_r_sdk_hyperdrive.rmd notebook to start the experiment runs in Azure ML. (You will not need to manually execute anything in the amiss_test_script.R script.)