Question about parsing replay files

[Gary7799]

Hi, I'm a computer science student at the University of Adelaide and I am working on a project which focuses on combining the game commentary transcript and replay statistics using GPT to improve the professional player’s efficiency.
I downloaded some replays and parsed them using S2Protocol(https://github.com/Blizzard/s2protocol) and found a lot of information but don't know what to use. I parsed game events but couldn't understand the meaning.

I would really appreciate it if you could let us know how to parse a replay file. For example, where could we find the data for units or buildings, like location, action, clusters, birthtime, dead time, attack/track, abilities, building type, building orders, etc.. Besides, there is still some raw data we can’t figure out the exact meaning of some attributes. We are very interested in knowing where we can find the complete documentation.
Thank you and this repo is so helpful to us who would like to try to analyze data from the SC2 game.

[StoicLoofah]

Cool project idea! I think ideally, this question should be in the https://github.com/ggtracker/sc2reader/discussions since it doesn't pertain directly to changes in the codebase, but happy to respond here first

There's a lot in the replay files. The best overview of the structures that are extracted via sc2reader are described in https://sc2reader.readthedocs.io/en/latest/ The structures are mapped pretty closely into sc2reader, so it definitely can help.

I myself mainly look at build orders, and that can largely be pulled out of the tracker events. You can see how I pulled some of that out at https://github.com/StoicLoofah/spawningtool/blob/7566f458b6b40cbdd58ee5cc8794c5b9adde71eb/spawningtool/parser.py#L319-L326

Hope that helps as a starting point. Let us know if you have more specific questions