Data dictionary and visualization inquiries

[patwater]

Hi thank you for sharing the excellent and intriguing paper! I had some questions about the AI output and other datasets after spending some time poking around and reviewing the readme. Apologies in advance if I'm missing something obvious!

• How does the "Answer" values in llama-70b.xlsx relate to the number of steps?

I.e. how do you get from this

-> to this

• What is the sample space for that llama-70b.xlsx dataset? I see there's 2034 rows there? It seems some of the munis are duplicated though with multiple questions provided for each municipality and there's 4091 records in the sample_enriched contextual dataset. That latter number makes more intuitive sense since there's ~3k counties in the US so that . Is the AI code parsing just for a subset of muni's / unincorporated areas? Where does the code classification for the unicorporated areas occur? That appears to be in the latter
• Is there a data dictionary for the questions i.e. what does 7, 12, 15 etc mean?

Also just to throw this out there and recognizing that time and resources aren't infinite but it'd be useful if there was a visualization tool (perhaps similar to https://tracktherecovery.org/ though not necessarily that shiny / built out) that enables folks to look at the results for a specific munipality / metro area / unincorporated areas therein...

Finally this type of AI classification seems like it'd be useful for revamping permitting protocols at a more operational level. See the screenshots at the end of this post for example: https://pioneeringspirit.xyz/how-harriss-housing-plan-could-work-a-bit-better

[dmilo75]

Thank you reading our work and for the interest in this project. Unfortunately the codebase reflects an earlier version of the project. Within the next few weeks we will have a new version up that reflects the paper you read. Apologies and thank you for your patience.