ardiawanbagus/sumobot-simulation-data

SumoBot Simulation Dataset

This dataset contains simulation data from SumoBot agent matches across various configurations.

Dataset Structure

Bot_A_vs_Bot_B/
└── Timer_X__ActInterval_Y__Round_Z__SkillLeft_W__SkillRight_V/
    └── *.parquet

File Naming Convention

Each parquet file corresponds to a specific matchup and configuration:

• Matchup: Bot_A_vs_Bot_B - The two agents competing
• Configuration: Simulation parameters
  • Timer: Match duration
  • ActInterval: Action interval (decision frequency)
  • Round: Round type (e.g., BestOf1, BestOf3)
  • SkillLeft: Left bot's special skill
  • SkillRight: Right bot's special skill

Loading the Dataset

Load All Data

from datasets import load_dataset

# Load entire dataset
ds = load_dataset("{repo_id}")

Load Specific Matchup

import glob
from datasets import Dataset
import polars as pl

# Load specific matchup parquet files
files = glob.glob("Bot_BT_vs_Bot_DQN/**/*.parquet", recursive=True)
df = pl.read_parquet(files)

Using with Polars/Pandas

import polars as pl
from huggingface_hub import hf_hub_download

# Download specific file
file_path = hf_hub_download(
    repo_id="{repo_id}",
    filename="Bot_BT_vs_Bot_DQN/Timer_15__ActInterval_0.1__Round_BestOf1__SkillLeft_Boost__SkillRight_Boost/Timer_15__ActInterval_0.1__Round_BestOf1__SkillLeft_Boost__SkillRight_Boost.parquet",
    repo_type="dataset"
)

# Read with Polars
df = pl.read_parquet(file_path)

Data Fields

Typical columns in the parquet files include:

• timestamp: Game timestamp
• bot_position: Left/Right
• action: Action taken by bot
• x_position, y_position: Bot coordinates
• velocity_x, velocity_y: Bot velocity
• collision_type: Type of collision event
• winner: Match outcome
• Additional game state variables

Use Cases

This dataset can be used for:

• Agent Evaluation: Compare performance across different bot architectures
• Configuration Analysis: Determine optimal game parameters
• Behavioral Analysis: Study agent strategies and movement patterns
• Reinforcement Learning: Train or fine-tune SumoBot agents
• Game Balance: Analyze competitive balance across configurations