Scoring server specs: multiprocessing support, start method, RAM, and CPU cores

[indie Cha]

Hi,

We have a few questions about the scoring server environment:

1. Is Python's multiprocessing module (e.g., ProcessPoolExecutor) supported during the training step?
2. If so, which start method is available — fork, spawn, or forkserver?
3. How much RAM is allocated per submission?
4. How many CPU cores are available?

Thank you!

[PhysioNet Challenge]

Hi! I’ll try to answer these as best I can.

1. Python’s multiprocessing is supported; I’ve seen people use it successfully in previous years.
2. It should support fork and forkserver.
3. 64GiB of RAM is allocated to the job, with 60GiB available for your code.
4. 16 CPU cores are available.

In general, we want to provide the teams with as much flexibility as possible for how they choose to solve the Challenge task, and we encourage teams to more efficiently use the available resources.

If you do submit something with multiprocessing that doesn’t work on our system, I’d be interested in figuring out why.

Best,

Audrey