Luke Winslow
This implementation is written in Python. This is a good fit for low-budget clients. It will be relatively easy to maintain by novice developers and did not take long to implement. It could run faster in other languages, but the data volume does not really warrant further work. On my laptop, it executes on >500MB datasets in under a minute.
The code should scale well to large datasets. The only scaling issue may be the memory footprint of keeping the full list of contribution values around for median calculation. But the data sizes just didn't seem to warrant a more complex, streaming median implementation. I am rusty with Python, so there may be some odd syntax choices. But I used this as a good excuse to dive back into the language.
- python (tested with v2.7)
- numpy (python package, install with
pip install numpy)
These are a few thoughts the client may want to consider going forward.
-
Ignoring out-of-order records implementation details could be interpreted more than one way. I implemented it as "ignore any previous year record". But out-of-order records within the same year are still counted in the "repeat donor" decision. These implementations have different tradeoffs.
-
In examining full datasets online, I found a number of negative donation values. Generally associated with "REATTRIBUTED" records. One may want to ignore or handle these records differently.
-
I didn't setup continuous integration (didn't want to push the limits with extra files in submission) but setting up Travis or something akin would make future development a bit easier.