Oracle

Oracle is an AI model that predicts match outcomes based on historical match data. Oracle is capable of predicting FRC match outcomes.

Description

The model is trained on a dataset that includes information from The Blue Alliance for basic match data, such as teams and points, and Statbotics for team performance evaluation metrics like the EPA model. The model implements the XGBoost algorithm to predict the score breakdown of a match.

Set up

1. Clone the repository:

   git clone [https://github.com/Cicchy/Oracle](https://github.com/Cicchy/Oracle)

2. Install dependencies:

   pip install pandas scikit-learn xgboost

How to use it?

You will find various models trained on the available dataset in the models folder. The models are classified in the following way:

• [dev]: Model still in development or highly experimental.
• [stable]: More finished and tested model.
• [release]: Official release used on the Oracle website.

We recommend you stick to the release version, but some stable versions should be fine for general use.

General match data

Match data fetched from The Blue Alliance used as a target for prediction.

• [year]: The current year the match occurred.
• [event_key]: The key assigned by the TBA for the event in which the match takes place.
• [comp_level]: The match level in the competition. [qm] for qualification matches, [qf] for quarter finals, [sf] for semi finals, [f] for finals.
• [match_number]: The number of the current match.
• [alliance_score]: The total score of the alliance.
• [alliance_auto_points]: The auto points scored by the alliance.
• [alliance_teleop_points]: The teleop points scored by the alliance.
• [alliance_endgame_points]: The endgame points scored by the alliance.
• [alliance_rp]: The ranking points earned by the alliance.

Team data

Individual team evaluation metrics, for better understanding of the EPA system, check out the EPA model.

• [team_epa]: The overall Expected Points Added (EPA) for the team.
• [team_epa_auto]: The EPA for the team during the autonomous period.
• [team_epa_teleop]: The EPA for the team during the teleoperated period.
• [team_epa_endgame]: The EPA for the team during the endgame period.
• [team_epa_trend]: The trend of the team's EPA over time.
• [team_win_rate]: The team's overall win rate.
• [team_win_streak]: The team's current win streak.

EPA data

• [epa_dif]: The difference in EPA between the red and blue alliances.
• [epa_ratio]: The ratio of the red alliance's EPA to the blue alliance's EPA.

Simple implementation

To use the model, load the trained model (or_dev_2018-2025v1.1.0.pkl or the latest release) and provide input data in the correct format. Here's a Python example:

import pickle
import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.impute import SimpleImputer
from xgboost import XGBClassifier

# Load the trained model
with open('models/or_dev_2018-2025v1.1.0.pkl', 'rb') as f:
    model = pickle.load(f)

# Define the features
features = [
    'year', 'event_key', 'comp_level', 'match_number', 'epa_diff', 'epa_ratio',
    'red_1_win_rate', 'red_2_win_rate', 'red_3_win_rate', 'blue_1_win_rate',
    'blue_2_win_rate', 'blue_3_win_rate', 'red_1_epa_trend', 'red_2_epa_trend',
    'red_3_epa_trend', 'blue_1_epa_trend', 'blue_2_epa_trend', 'blue_3_epa_trend',
    'red_auto_points', 'blue_auto_points', 'red_teleop_points', 'blue_teleop_points'
]

# Function to predict probabilities
def predict_win_probability(input_data):
    input_df = pd.DataFrame([input_data], columns=features)
    input_df = pd.get_dummies(input_df, columns=['event_key', 'comp_level'], drop_first=True)
    # Reindex columns to match training data
    input_df = input_df.reindex(columns=X_train.columns, fill_value=0)
    imputer = SimpleImputer(strategy='mean')
    input_imputed = imputer.transform(input_df)
    scaler = StandardScaler()
    input_scaled = scaler.transform(input_imputed)
    win_probability = model.predict_proba(input_scaled)[:, 1][0]
    return win_probability

# Example input data
input_data = {
    'year': 2018, 'event_key': '2018abca', 'comp_level': 'qm', 'match_number': 6,
    'epa_diff': 100, 'epa_ratio': 1.2, 'red_1_win_rate': 0.8, 'red_2_win_rate': 0.7,
    'red_3_win_rate': 0.9, 'blue_1_win_rate': 0.6, 'blue_2_win_rate': 0.5,
    'blue_3_win_rate': 0.8, 'red_1_epa_trend': 0.1, 'red_2_epa_trend': -0.2,
    'red_3_epa_trend': 0.3, 'blue_1_epa_trend': -0.1, 'blue_2_epa_trend': 0.2,
    'blue_3_epa_trend': -0.3, 'red_auto_points': 20, 'blue_auto_points': 15,
    'red_teleop_points': 100, 'blue_teleop_points': 90
}

# Make predictions
win_probability = predict_win_probability(input_data)
print(f'Probability of Red Alliance Winning: {win_probability}')