metadata
license: mit
task_categories:
- reinforcement-learning
- tabular-classification
- text-classification
- tabular-to-text
tags:
- decision-making
- reinforcement-learning
- games
- rpg
- RL
- gym
- human-gameplay
task_ids:
- emoji_quest_rl
pretty_name: Emoji Quest RL Dataset
dataset_info:
configs:
- config_name: game_states
data_files:
- split: train
path: game_states.csv
features:
- name: timestamp
dtype: int64
- name: action
dtype: string
- name: reward
dtype: float64
- name: outcome
dtype: string
- name: decision_type
dtype: string
- name: player_state_snapshot
dtype: string
- name: enemy_info
dtype: string
- name: player_state
dtype: string
- name: enemy_state
dtype: string
- name: enemy_count
dtype: int64
- name: game_state
dtype: string
- config_name: decisions
data_files:
- split: train
path: decisions.csv
features:
- name: timestamp
dtype: int64
- name: action
dtype: string
- name: reward
dtype: float64
- name: outcome
dtype: string
- name: decision_type
dtype: string
- name: player_state_snapshot
dtype: string
- name: enemy_info
dtype: string
- name: player_state
dtype: string
- name: enemy_state
dtype: string
- name: enemy_count
dtype: int64
- name: game_state
dtype: string
default: game_states
___________ __.__ ________ __
\_ _____/ _____ ____ |__|__| \_____ \ __ __ ____ _______/ |_
| __)_ / \ / _ \ | | | / / \ \| | \_/ __ \ / ___/\ __\
| \ Y Y ( <_> ) | | | / \_/. \ | /\ ___/ \___ \ | |
/_______ /__|_| /\____/\__| |__| \_____\ \_/____/ \___ >____ > |__|
\/ \/ \______| \__> \/ \/
Emoji Quest RL Dataset
Reinforcement learning dataset from Emoji Quest RL gameplay. Single human playthrough of Emoji Quest RL — turn-based emoji RPG with combat, stat upgrades, resting & crafting. The game is available for download in the /gym/ folder of the repo so players can create their own datasets.
It contains 85 decision points / states → full session: lvl 1→8, 13 enemies defeated, 1451 gold earned, ~169s playtime. Two splits / files (Parquet + CSV):
- game_states — snapshots at every decision point (default)
- decisions — focused on player choices + immediate results
Key Columns (all rows):
- timestamp — event time
- action — what happened (e.g. "Spell", "enemy_attack", "upgrade_int", "inn_rest", "craft_armor_1")
- reward — shaped signal (-100..500): + for damage/victory/progression, - for taken damage
- outcome — result type (damage_dealt, damage_taken, victory, progression, etc.)
- decision_type — category (Spell, enemy_attack, upgrade_stat, rest_inn, craft_armor)
- player_state / player_state_snapshot — JSON: level, hp/mp, str/int/vit/def, gold, skillPoints, exp, equipment flags
- enemy_info / enemy_state — JSON: emoji enemy name (💾 Datavore etc.), hp, attack, boss/eliteness
- enemy_count — 1–13
- game_state — always "playing"
Files
game_states.csv
Game state snapshots at each decision point. Contains player and enemy stats.
decisions.csv
Player decisions with rewards and outcomes. Contains action choices and results.
Summary
- Total states: 85
- Total decisions: 85
- Total enemies defeated: 13
- Total gold earned: 1451
- Total XP earned: 121
- Play time: 169 seconds
- Final level: 8
- Final gold: 1451
- Final HP: 210
- Final MP: 20