How jb250101/info8665--airplane works
Main Components
a. Data Sources and Data Management:
• The “data/” directory holds datasets (Excel files, CSVs) and a database script. This indicates that raw and preprocessed data is maintained separately.
• The “docs/” directory—which, although it holds documentation, confirms the presence of a structured process and stakeholder communication.
b. Machine Learning Services:
• The “dev/” directory includes various Python modules for data collection, preprocessing, training, machine learning services, and predictions.
• The README explicitly describes three core Flask-based microservices:
• Preprocessing Service (handles data cleaning/encoding, accessible at port 5002)
• Training Service (trains the flight fare prediction model, accessible at port 5003)
• Prediction Service (returns flight fare predictions based on processed inputs, accessible at port 5004)
c. Model Artifacts:
• The “models/” directory contains the trained model (flight_fare_model.pkl), various encoders (for airline, destination, etc.), and a scaler. This is your ML model repository.
Visualizing Relationships and Interactions
• The Training Service uses processed data to train the model, then saves the resulting artifacts into the “models/” repository.
• The Prediction Service uses the trained model (by loading from “models/”) to generate predictions based on incoming requests with already-preprocessed data.
Architectural Patterns and Design Principles
• Separation of Concerns: Data ingestion, data transformation, model training, and inference are isolated in their respective components.
• MLOps Pipeline: While not extremely detailed in the file tree, the README and structure imply integration of CI/CD workflows using GitHub and Azure Boards for deployment and collaboration. You could represent this as an external “CI/CD Pipeline” component that interacts with the code repository and deployment environment.
Final Elements
• Each component (Preprocessing Service, Training Service, Prediction Service, Data Storage, Model Repository, CI/CD pipeline) should be clearly labeled.
• Adding components or annotations that indicate where and how each service is deployed (e.g., noting the local ports: 5002, 5003, and 5004).