Datasetpublic

Quazim0t0/Warcraft-Simulations-5x

Warcraft I — AI vs AI Simulation Dataset Rust AI-vs-AI Warcraft I engine that produces deterministic game replays as structured data suitable for ML research, more so for perdictive world model training. Dataset stats Stat Value Games 5 Total tick rows 61,030 ticks Ticks per second (in-game) 20 Quick start from datasets import load_dataset ds = load_dataset("parquet", data_files="*.parquet", split="train") # Economy over time… See the full description on the dataset page: https://huggingface.co/datasets/Quazim0t0/Warcraft-Simulations-5x.

sourceHugging Faceupdated 4mo agoView on Hugging Face

0likes12downloads

Dataset Card

Warcraft I — AI vs AI Simulation Dataset

Rust AI-vs-AI Warcraft I engine that produces deterministic game replays as structured data suitable for ML research, more so for perdictive world model training.

Dataset stats

Stat	Value
Games	5
Total tick rows	61,030 ticks
Ticks per second (in-game)	20

Quick start

python

from datasets import load_dataset

ds = load_dataset("parquet", data_files="*.parquet", split="train")

# Economy over time for the human player
import pandas as pd
df = ds.to_pandas()
print(df[["tick", "humans"]].head())

Row schema

Each row is one simulation tick.

Column	Type	Description
`seed`	int64	RNG seed (reproduces the exact game)
`winner_id`	int64?	0 = Humans, 1 = Orcs, null = unresolved
`tick`	int64	Tick number (starts at 1, 20 ticks/sec)
`game_time_seconds`	float64	Elapsed in-game time
`humans`	struct	Human player state + entities (see below)
`orcs`	struct	Orc player state + entities (see below)
`neutral_entities`	list\\<struct\\>	Gold mines
`events`	list\\<struct\\>	All game events this tick
`screenshot`	string?	Relative path to PNG (present without `--embed-images`)
`image`	Image	Embedded PNG, viewable in HF dataset viewer (only with `--embed-images`)

Player struct (`humans` / `orcs`)

Field	Type	Description
`gold`	int32	Current gold
`lumber`	int32	Current lumber
`food_used`	int32	Food consumed
`food_cap`	int32	Food capacity
`apm`	float64	Actions per minute (60 s rolling window)
`total_actions`	int64	Cumulative AI orders issued
`unit_count`	int64	Living combat units
`building_count`	int64	Standing buildings
`worker_count`	int64	Living workers
`is_eliminated`	bool	Player eliminated
`entities`	list\\<struct\\>	All owned non-forest entities

Entity struct

Field	Type	Description
`id`	int32	Unique entity ID
`entity_type`	string	e.g. `"Footman"`, `"TownhallHumans"`
`x` / `y`	float32	World position (pixels)
`tile_x` / `tile_y`	int32	Tile position (64×64 map)
`hp` / `max_hp`	int16	Hit points
`hp_pct`	float32	`hp / max_hp`
`state`	string	AI state (`"Idle"`, `"Attacking"`, …)
`target_id`	int32?	Attack target entity ID
`gold_reserve`	int32?	Remaining gold (gold mines only)

Event types (`events[].type`)

Attack, Death, Move, ResourceGain, UnitTrained, BuildingConstructed, AIOrder, GameOver

Units

Side	Workers	Tier 1	Tier 2	Tier 3
Humans	Peasant	Footman	Archer	Knight
Orcs	Peon	Grunt	Spearman	Raider

Both sides can also build Catapults.

Tech tree buildings

Farm → Barracks → LumberMill → Stable / Blacksmith

Citation

bibtex

@misc{{wc1-dataset,
  title   = {{Warcraft Simulation Dataset}},
  year    = {{2026}},
  note    = {{Created by Dean Byrne}},
}}