Speed up process_map function by using data.table

[huisman]

The process_map function felt a bit slow, so I rewrote parts of it to try and speed it up. Based on the profvis profiling output, I replaced the dplyr style slicing with data.table methods, but kept the output exactly the same (column order, sorting).

I used microbenchmark to runs process_map(data) 50 times and report how much less time it took (median time) to run the new version compared to the old. The resulting speed up is dependent on the data set:

• sepsis takes ~33% less time
• patients takes ~34% less time
• hospital_billings takes ~70% less time (from 1525 ms to 459ms)
• traffic_fines takes ~72% less time (from 1457ms to 395ms)

On my real life dataset containing almost 200,000 rows it takes ~30% less time to run process_map(data, frequency('absolute-case')) (from 32s to 21s). The benchmark on the real life dataset was run on a different device than the example datasets.

[gertjanssenswillen]

Thanks for the updates. I'm doing some checks on dev, merge with master and publish on CRAN. (also edeaR later).
As a token of appreciation, I can add you as a contributor with your github username, or if your actual name if you let me know.

[huisman]

Thanks for the updates. I'm doing some checks on dev, merge with master and publish on CRAN. (also edeaR later). As a token of appreciation, I can add you as a contributor with your github username, or if your actual name if you let me know.

Thanks, no need to add me as a contributor, getting these changed merged is appreciation enough :)