How tanthinhdt/cbbl-2025 works
Project type and purpose
• This is not a web‐app or microservice suite but rather a machine‐learning experimentation/benchmarking project.
• It houses Jupyter notebooks for binary/multi‐class classification and continuous‐value regression, evaluates multiple model families (decision trees, random forests, gradient boosting, logistic/linear regression, TabPFN variants, GPT‐style models), and captures experiment outputs under various preprocessing strategies.
File structure
• Top‐level:
• classification/ ← Jupyter notebooks for classification experiments
• regression/ ← Jupyter notebooks for regression experiments
• results/ ← CSV outputs organized by preprocessing category
• requirements.txt ← Python dependencies list
• Under results/ you have subfolders for each preprocessing axis (missing_values, feature_scaling, feature_encoding, dimensionality_reduction, outliers) plus a “main” folder for final model comparisons.
• There are no dedicated Python modules or services—experiments are driven directly from notebooks.
README
• README.md only contains the title (“# cbbl‐2025”). No architecture hints, so we infer from the directory layout that the core architecture is a data→preprocessing→modeling→results pipeline orchestrated by notebooks.
a. Main components
• Raw Data Sources (train.csv, test.csv under classification/ and regression/)
• Preprocessing Modules
• Missing‐Value Handler
• Feature Scaling
• Feature Encoding
• Outlier Detection/Handling
• Dimensionality Reduction
• Experiment Orchestration Layer (Jupyter Notebooks)
• classification notebooks
• regression notebooks
• Model Training & Evaluation Engines
• Decision Tree / Random Forest / Gradient Boosting
• (Logistic/Linear) Regression
• TabPFN v2 (standard, optimized, exp)
• GPT-style (BioGPT, GPT2, LLaMA3)
• Results Storage (results/ subfolders)
• Visualization & Reporting (images/ plus any embedded notebook charts)
b. Relationships & Data Flow
• From each Preprocessing Module → Experiment Orchestrator (classification/regression notebooks)
• Within notebooks, branch to multiple Model Engines in parallel
• Model Engines output metrics/CSV → Results Storage
• Final aggregation → Visualization layer
c. Architectural Patterns & Principles
• Modular pipeline: each preprocessing step is a reusable component
• Parallel model evaluation: notebooks spin up multiple algorithms concurrently
• Separation of concerns: data preparation, model training, and result collection are clearly partitioned
d. Key Technologies & Frameworks
• Python 3.x (see requirements.txt)
• Jupyter Notebooks as orchestrators
• scikit-learn, TabPFN library, Hugging Face Transformers (for GPT variants)
• pandas for data handling, matplotlib/seaborn for plots