DecryptX Helper Library

A Python library for the DecryptX Round 3 Data Cleaning Contest.

Installation

Install directly from GitHub:

pip install git+https://github.com/keanesc/decryptx-helper.git

Or for development:

git clone https://github.com/keanesc/decryptx-helper.git
cd decryptx-helper
pip install -e .

Dataset Access

The library automatically downloads the FIFA dataset from the DecryptX server when you call load_data(). The dataset is cached locally in ~/.cache/decryptx/ and refreshed every 24 hours.

Environment Variables (Optional)

You can customize the dataset location using an environment variable:

• DECRYPTX_DATA_PATH: Direct path to a local dataset file (bypasses download)

Example:

# Use a local file (no download)
export DECRYPTX_DATA_PATH="/path/to/your/fifa_raw_data.csv"

Google Colab Note: The dataset will be automatically downloaded on first use. No manual upload required!

Quick Start

from decryptx import login, load_data, submit

# 1. Login with your team credentials
session = login(team_name="YourTeamName", password="your_password")

# 2. Load the raw FIFA dataset
df = load_data()

# 3. Clean your data (YOUR WORK GOES HERE)
df_clean = your_cleaning_function(df)

# 4. Submit your cleaned dataset
# This will automatically split data, train a fixed model, evaluate performance, and submit the score.
result = submit(session, df_clean)
print(f"Remaining attempts: {result['remainingAttempts']}/5")

API Reference

login(team_name: str, password: str) -> dict

Authenticate with the DecryptX server.

Returns: Session dictionary containing teamId, sessionId, and qualification status.

load_data() -> pd.DataFrame

Load the raw FIFA dataset that needs cleaning.

Returns: pandas DataFrame with the raw data.

submit(session: dict, df: pd.DataFrame) -> dict

Submit your cleaned dataset for evaluation.

This function:

1. Splits your data into train/test sets (fixed random seed)
2. Trains a standardized Random Forest model
3. Evaluates RMSE on the test set
4. Submits the score to the leaderboard

Returns: Submission result dictionary.

Submit your score to the leaderboard.

Returns: Submission result with remainingAttempts and status.

Rules

1. Fixed Parameters: The train/test split uses random_state=42 and test_size=0.2. Do not modify these.

2. 5 Submission Limit: You have exactly 5 submission attempts total (lifetime limit).

3. 1 Minute Cooldown: Wait at least 1 minute between submissions.

4. RMSE Scoring: Lower RMSE is better. The target is the player's Overall Rating (OVA).

5. Data Cleaning Focus: The competition is about data cleaning, not model architecture. A simple model on well-cleaned data often beats a complex model on dirty data.

Tips

• Handle missing values appropriately
• Parse numeric values from strings (e.g., "€103.5M" → 103500000)
• Handle height/weight formats (e.g., "170cm" → 170)
• Remove or encode special characters
• Consider feature engineering from the available columns

License

MIT License