mcp_server/README.md

MCP Server: Monte Carlo Simulation with a TFX Pipeline

This directory contains the mcp_server, a tool for running Monte Carlo simulations for race strategy, now enhanced with a production-grade MLOps workflow using TensorFlow Extended (TFX).

Overview

The server provides race strategy insights by running simulations based on models for tire degradation, fuel consumption, and race pace. The ML models are now managed by a TFX pipeline that automates the process of data preparation, training, evaluation, and deployment.

Environment Setup

IMPORTANT: This project uses the tfx library, which currently requires Python 3.11 or lower. Please ensure you are using a compatible Python version to run the pipeline and tests.

1. Create a Virtual Environment:

   python3.11 -m venv venv
   source venv/bin/activate
   

2. Install Dependencies: First, install TFX, Apache Beam, and other core dependencies:

   pip install "tfx>=1.15.0" "apache-beam[interactive]>=2.46.0" pytest
   

   Then, install the project-specific dependencies:

   pip install -e .
   

MLOps Workflow

The ML model lifecycle is managed by a series of scripts that should be run in order.

1. Prepare Data

This script downloads the raw race data, processes it into a unified CSV file, and places it in the correct directory for the TFX pipeline to use.

python3 mcp_server/prepare_data.py

2. Run the TFX Pipeline

This script orchestrates the end-to-end ML workflow. It will ingest the prepared data, train the models, evaluate their performance, and (if they pass a quality threshold) deploy them to the serving directory.

python3 mcp_server/pipeline/pipeline.py

3. Visualize Model Performance

After the pipeline has run, you can generate a visual report of the model performance metrics. This script reads the latest evaluation_metrics.json and creates a bar chart saved as evaluation_results.png in the latest model's directory.

python3 mcp_server/visualize_metrics.py

4. Start the Server

Once the pipeline has successfully run and a model has been deployed, you can start the MCP server. The server will automatically load the latest blessed model from the pipeline.

python3 mcp_server/main.py

Testing

To verify the correctness of the pipeline components, run the unit tests using pytest. Make sure you are in the compatible Python 3.11 environment with all dependencies installed.

pytest mcp_server/tests/