How albertbs2/predict-bikes-availability works
Main Components
• Data Layer: Represent the raw and preprocessed data housed under the “data/” directory. This includes the different CSV files for training, test, and preprocessed datasets.
• Libraries & Utilities: Under “libs/”, include a component for utility functions, feature engineering, and data loading. These are the reusable functions called by different notebooks.
• Exploration & Analysis: Mark the “data_analysis.ipynb” notebook as the module responsible for exploratory data analysis and visualization.
• Model Training & Tuning: Create nodes for each model’s notebook found in “models/” (e.g., Logistic Regression, KNN, Random Forest, XGBoost, CatBoost). Emphasize that these notebooks handle training, testing, and hyperparameter tuning.
• Output & Best Parameters: Represent the “output/” sub-directory, especially “best_params/”, where JSON files store optimum parameters for each model.
• Final Results: Include “main.ipynb” as the component that uses the tuned models (especially the winning Random Forest) to generate final predictions and graphs.
Design the Relationships and Interactions
• Feature Engineering & Preprocessing: Illustrate a connection from the data layer to the exploration analysis notebook (data_analysis.ipynb) that uses utility functions from “libs/”.
• Final Integration: Finally, indicate how “main.ipynb” pulls together the cleaned data, tuned parameters (especially for the Random Forest model), and possibly visual outputs from the analysis, to generate final predictions/graphs.
Architectural Patterns and Design Principles
• Experimentation Workflow: Highlight the iterative nature of machine learning experimentation—from data analysis through multiple model training notebooks, hyperparameter tuning, and culminating in final production usage in “main.ipynb”.