- Spend 1-2 days revisiting the workplan after working for one quarter, when you know the code better. Then you'll be able to set PI expectations appropriately for what can be accomplished.
- Provide examples that a PI can integrate into their own code. Offer to help, but often the examples are very instructional and it helps for them (or their surrogates) to actually do the code changes
- Keep in mind the size of the PI's research group when making suggestions. While GPUs might speed the code, if the PI doesn't have a large group that can maintain the changes, it might not be smart to recommend that strategy
- Keep in mind the PI's goals. Sometimes speeding up the code requires using many more processors and may limit what the PI can do with their given allocation. While the parallelization work has likely set the PI on the right path for requesting larger allocations in the future and they can request supplements, they may have immediate goals to accomplish with the current allocation size.
GIS and R tips