Running on Theta

Note

The examples so far assume that your DeepHyper models run in the same Python environment as DeepHyper and each model runs on a single node. If you need more control over model execution, say, to run containerized models, or to run data-parallel model training with Horovod, you can hook into the Balsam job controller. See Configuring model execution with Balsam for a detailed example.

Note

Also make sure not to change any of our workflow names or directory structures as this is crucial for accurate node utilization assessments when using deephyper-analytics.