Answering Questions from the Facts

gem reports what the replay records and leaves the interpretation to you. These recipes show what that looks like in practice: each answers one question a player or analyst might ask, using gem.parse, the facts on ParsedMatch and pandas. None of them uses a tag, score or verdict, and none of them uses anything deprecated in 0.12.

Question	Facts it joins
Did the Roshan team win the next fight?	Roshan kills, Aegis events, fights, the game clock
How fast did each core farm from 10 to 20 minutes?	Minute curves, neutral deaths in the combat log, positions and `gem.region_of`
How often did a smoke lead to a kill?	`gem.build_smoke_analysis`, hero deaths in the combat log, fights
Where did each team ward, and what could the wards see?	Wards, `gem.region_of`, hero positions and the replay's own visibility
Where were the fights, and who got the kills?	gem's fights, buildings and Roshan kills, the game clock
When did each objective fall, and did teams convert their edges?	Buildings, Roshan and Tormentor kills, building damage in the combat log, fights
What moved the gold and XP lead?	The gold ledger, total earned gold and XP, XP and deaths in the combat log
How did each lane go?	Hero positions on OpenDota's lane grid, minute curves, the gold ledger, deaths and teleports in the combat log
Who took each rune, and what happened while a power rune lasted?	`match.runes`, `PICKUP_RUNE` entries, rune buffs, kills and damage in the combat log, buildings and Roshan kills

How the recipes are laid out

Each recipe is a short script in examples/cookbook/. It has one function that takes a ParsedMatch and returns a DataFrame with fixed columns, and a main that runs it on one replay or on many:

python examples/cookbook/roshan_next_fight.py match.dem
python examples/cookbook/roshan_next_fight.py replays/*.dem

With more than one replay, main parses them in parallel with gem.parse_many. Its worker processes re-import the script, so code that calls parse_many needs an if __name__ == "__main__": guard; a script piped in through standard input cannot use it.

The functions are tested in tests/test_cookbook.py, both on small built matches and on a fixture replay, so the code on these pages runs as shown.

The sample

The outputs come from the TI match 8974053011 (MOUZ vs Natus Vincere; the wards, fights, objectives, lead, lanes and runes recipes use 8856501050, the home page's match) and, for the totals, the 9 local fixture replays (tests/fixtures/opendota/). Nine pro matches are enough to show the method, not to draw conclusions; run the same code over your own replays.

Times

Replay ticks keep running during pauses, and the in-game clock does not. The recipes measure every delay and window with the match's game clock (match.game_clock), so a pause between two events is not counted.