Skip to main content

​
Overview

The /cruncher directory contains example model implementations from the perspective of Crunchers - the ML engineers who submit competing models to Crunch Protocol prediction challenges. This directory demonstrates how models are structured, packaged, and submitted within the Crunch Protocol Ecosystem.

​
Core Components

​
1. Model Package (model_package/)

Provides the standard interface that all models in this Crunch must implement. Model implementation example: 
from crunchdao.crunch_example.iris import IrisModelBase

class MyIrisModel(IrisModelBase):
    def train(self, train_data: pd.DataFrame) -> None:
        """Train the model on provided training data"""
        pass
    
    def infer(self, dataframe: pd.DataFrame) -> pd.DataFrame:
        """Generate predictions for test data"""
        predictions = self.predict(dataframe)
        return pd.DataFrame({'prediction': predictions})

​
2. Model Implementations (models/)

This directory includes three complete model implementations, each showcasing a different ML approach. As a Coordinator, you can use them to test your node or share them with Crunchers as references — either to guide their own implementations or serve as benchmark models for pre-Crunch evaluation.

​
Integration with Coordinator

The models integrate with the Coordinator through:
  1. Model Orchestration: Coordinator sends training data via gRPC
  2. Containerized Execution: Models run in isolated Docker containers
  3. Prediction Collection: Coordinator collects predictions from all models
  4. Performance Scoring: Coordinator evaluates model performance

Coordinator Node Setup

In the next section we will walkthrough the core components to set up your Coordinator node.