Boosting Scoring Throughput

By default, the SCORE.sh script calculates all metrics (“scores”) for all plans in an input ensemble. When the ensemble is large, this can take a long time.

You can use a combination of two techniques to increase scoring throughput:

1. “Shard” the ensemble of plans into files with fewer plans, e.g., divide it into 10 smaller files, and score the shards in parallel.
2. Score independent categories of metrics—“general”, “partisan”, “minority”, “compactness”, and “splitting”—separately and in parallel.

When used together, you can substantially boost throughput.

Used together, the basic process is as follows:

• Shard the ensemble of plans, e.g., into 10.
• Score each category of metrics for each shard separately, using the --mode arg on SCORE.sh.
• “Vertically” concatenate each set of shards back into one file of scores -or- by-district measurements for each category, and
• Optionally, “horiztonally” join all the categories files into one overall scores or by-district file.

There are three utility bash scripts in the scripts/throughput directory to support this:

• SHARD.sh – e.g., scripts/throughput/SHARD.sh /path/to/plans.jsonl
• CONCAT_FILES.sh – scripts/throughput/CONCAT_FILES.sh /path/to/csvs "NC_congress_scores_compactness_*.csv" or scripts/CONCAT_FILES.sh /path/to/csvs "NC_congress_by-district_compactness_*.jsonl" --no-header
• JOIN_CSVS.sh – e.g., scripts/throughput/JOIN_CSVS.sh /path/to/csvs "NC_congress_scores_*.csv"

The SCORE.sh script takes an optional --mode argument to specify which category of metrics to calculate. The default is to calculate all categories.