Model Training and Validation
Dr. Smith has reached the most critical step in her mission to build a pioneering pneumonia prediction model: training the model. Before Dr. Smith can train her model, she will need to split both her and Dr. Jones' Datasets into two: a training Dataset and a testing Dataset. To perform such a feat, Dr. Smith needs to develop a Python container that can divide the Datasets based on her pre-defined splitting criteria. After completing her container, she is going to push it to her Workgroup's ECR and then create new Generalized Compute Code using her container image. She then can run the Code over both her and Dr. Jones' data, leaving her with two training Datasets and two testing Datasets (a set of testing and training Datasets on Dr. Jones' Rhino Client and a set on Dr. Smith's Rhino Client).
With training and testing Datasets ready to go, Dr. Smith is ready to start developing the final container that is going to be responsible for the federated training and validation of her pneumonia model. Dr. Smith will use the NVFlare framework while developing her container. Once more, she will push the completed container to her ECR, but this time she must create an NVFlare Code Object to run the training and validation over both Dr. Jones' and Dr. Smith's Datasets.
Help Dr. Smith split her and Dr. Jones data into training and testing Datasets using Generalized Compute Code. Then help Dr. Smith use NVFlare to build her training and validation workflow for her federated pneumonia prediction model. Finally, help her with the most crucial part of her project: running the federated training and validation of the model using both Dr. Jones' Dataset and Dr. Smith's Dataset by completing Notebook #4: Running Federated Training of the Pneumonia Model
Supplemental Resources
- Pushing Containers to the ECR
- What is a Generalized Compute Code Object?
- Creating a New Generalized Compute Code Object
- Running a Generalized Compute Code Object
- What is NVFlare?
- Creating New NVFlare Code or Code Version
- Running an NVFlare Model
If you need support at any time, feel free to contact support@rhinohealth.com.