Module 2: Classification & Production Pipelines 🎯

From Notebooks to Professional ML Systems

📍 Location: 2_Classification/
🎯 Prerequisite: Module 1: Regression Mastery
➡️ Next Module: Module 3: Neural Networks from Scratch

Welcome to the Classification Module of SAIR – where you transition from experimental notebooks to production-ready ML systems with professional pipelines and deployment architecture.

🎯 Is This Module For You?

✅ Complete this module if:

You've mastered regression and want to tackle classification problems
You're ready to build professional ML pipelines
You want to learn industry best practices for ML systems
You're preparing for ML engineering roles

🚀 Review and continue if you're experienced:

You've built classification models but want production experience
You're familiar with sklearn but want pipeline architecture skills
You want to add MLflow and modular design to your toolkit

🛠️ Tools You'll Master

These production tools transform your ML code from experiments to enterprise-ready systems.

📚 What You'll Learn

Lecture	Focus	Time Estimate	Mastery Level
`Lecture_4.ipynb`	Classification from Scratch	4-5 hours	Essential
`Lecture_5.ipynb`	Production Pipeline System	5-6 hours	Professional
`Pipeline/` System	Modular Architecture	6-8 hours	Industry Ready

🗺️ Your Learning Journey

Phase 1: Algorithm Fundamentals 🎯

Start with: Lecture_4.ipynb

Implement logistic regression from first principles
Understand classification metrics and evaluation
Build intuition for decision boundaries and probability

Phase 2: Pipeline Development 🚀

Continue with: Lecture_5.ipynb

Transform notebooks into modular code
Learn configuration management
Set up experiment tracking and hyperparameter tuning

Phase 3: Production Architecture 📚

Master with: Pipeline/ system

Build end-to-end ML pipeline
Implement professional project structure
Deploy with Streamlit applications

🎯 Learning Outcomes

After completing this module, you will be able to:

Skill	Where You Build It
Implement logistic regression from scratch	`Lecture_4.ipynb`
Interpret classification metrics: precision, recall, F1, ROC-AUC	`Lecture_4.ipynb`
Transform a notebook into a modular ML codebase	`Lecture_5.ipynb`
Design a configuration-driven pipeline	`Pipeline/`
Run grid search and hyperparameter tuning at scale	`Pipeline/models/`
Compare experiments in MLflow and select the best model	`Pipeline/`
Deploy a classifier as a Streamlit web app	`Pipeline/streamlit_app/`

💡 Our Learning Philosophy

"From experimental code to production systems."

At SAIR, we believe modular, maintainable code separates hobby projects from professional systems. This module teaches you to architect ML solutions that scale and can be maintained by teams.

This is where you become an ML engineer, not just a model builder.

🚀 Quick Start Guide

For Sequential Learners:

# 1. Start with classification fundamentals
jupyter notebook Lecture_4.ipynb

# 2. Learn pipeline transformation
jupyter notebook Lecture_5.ipynb

# 3. Explore the production pipeline
cd Pipeline
python run_pipeline.py

For Pipeline-Focused Learners:

# Dive directly into professional architecture
cd Pipeline
python run_pipeline.py

# Run the Streamlit app
uv run streamlit run streamlit_app/app.py

Run the Complete Example:

# Test the breast cancer pipeline
python breast_cancer_pipline.py

🏗️ Professional Pipeline Architecture

🚀 Spaceship Titanic ML Pipeline Example

The Pipeline/ directory contains a complete, production-ready ML system that transforms Lecture 5 concepts into a professional codebase.

Key Features:

✅ Modular Architecture - Separate data, models, config, utils
✅ Advanced Feature Engineering - Custom transformers for domain-specific features
✅ Multi-Model Training - 7+ algorithms with systematic comparison
✅ Hyperparameter Tuning - Cross-validation and optimization
✅ MLflow Experiment Tracking - Reproducible experiments
✅ Streamlit Deployment - Interactive web application

Pipeline Structure:

Pipeline/
├── config/              # Configuration Management
│   ├── config.py        # Centralized settings and paths
│   └── __init__.py
├── data/                # Data Processing
│   ├── load_data.py     # Data ingestion and splitting
│   ├── preprocessing.py # Cleaning & preparation pipelines
│   ├── feature_engineering.py # Custom feature creation
│   └── raw/            # Source datasets
├── models/              # ML Modeling
│   ├── base_model.py    # Abstract base classes
│   ├── train_model.py   # Training orchestration
│   ├── evaluate_model.py # Comprehensive evaluation
│   └── hyperparameter_tuning.py # Systematic optimization
├── utils/               # Shared Utilities
│   └── mlflow_utils.py  # Experiment tracking helpers
├── streamlit_app/       # Deployment
│   └── app.py          # Web interface for predictions
└── run_pipeline.py      # Main execution script

Run the Complete Pipeline:

cd Pipeline

# Execute full pipeline
python run_pipeline.py --mode full

# Or run specific steps
python run_pipeline.py --mode preprocessing    # Data only
python run_pipeline.py --mode training        # Models only  
python run_pipeline.py --mode evaluation      # Evaluation only

View MLflow Experiments:

mlflow ui --backend-store-uri mlruns
# Open http://localhost:5000 in your browser

🎯 Capstone Project: Build Your Pipeline

Your Mission:

Apply the pipeline architecture to a classification problem of your choice, inspired by the Spaceship Titanic example.

Success Criteria:

✅ Implement modular pipeline structure
✅ Advanced feature engineering for your domain
✅ Multi-model comparison and selection
✅ MLflow experiment tracking
✅ Streamlit deployment interface
✅ Professional documentation

Project Ideas (Inspired by Spaceship Titanic):

🏥 Medical Diagnosis - Patient outcome prediction
💳 Fraud Detection - Transaction classification
📧 Spam Filter - Email categorization system
🛒 Customer Churn - Retention prediction
🎯 Sentiment Analysis - Review classification
🚀 Custom Dataset - Your own classification problem!

Follow the Pattern:

Study the Pipeline/ structure and adapt it for your project:

Replace dataset loading in data/load_data.py
Customize feature engineering in data/feature_engineering.py
Modify model portfolio in models/base_model.py
Update the Streamlit app for your domain

🌟 Student Inspiration: Spaceship Titanic Pipeline

The included Pipeline/ demonstrates exactly what you'll build:

What Makes It Professional:

Configuration Management: Centralized settings in config.py
Feature Engineering: Custom SpaceshipFeatureEngineer class
Model Portfolio: 7+ algorithms with hyperparameter tuning
Experiment Tracking: MLflow for reproducibility
Modular Design: Each component independently testable

Key Learning Outcomes:

After studying this pipeline, you'll be able to: ✅ Build modular ML pipelines from scratch
✅ Implement domain-specific feature engineering
✅ Compare multiple models systematically
✅ Track experiments with MLflow
✅ Create reproducible research
✅ Structure projects for collaboration

Adaptation Guide:

# In your project, replace Spaceship Titanic specifics:
# data/load_data.py → Your dataset loading
# data/feature_engineering.py → Your domain features  
# models/base_model.py → Your model portfolio
# streamlit_app/app.py → Your application interface

📊 Reading MLflow Results

When you run the pipeline, MLflow logs every experiment. Here's how to interpret the results:

# Open the MLflow dashboard
mlflow ui --backend-store-uri Pipeline/mlruns
# Then open: http://localhost:5000

Metric	What It Means	Good Range
`val_accuracy`	Accuracy on held-out validation set	>85% for most problems
`val_f1`	Harmonic mean of precision and recall	>0.80 for balanced classes
`val_roc_auc`	Area under the ROC curve	>0.85 is strong
`train_accuracy` vs `val_accuracy`	Gap > 10% = overfitting	Keep gap under 5%

Model selection rule: pick the model with the best val_f1, not train_accuracy. High training accuracy with low val accuracy = overfitting — try more regularization or less complexity.

🔧 Troubleshooting

Problem	Likely Cause	Fix
`mlflow.exceptions.MlflowException`	Tracking dir not found	Run pipeline from the `Pipeline/` directory
Streamlit app shows stale predictions	Old model pickle loaded	Delete `models/` and re-run `--mode full`
`KeyError` in feature engineering	Column renamed in your dataset	Update column names in `config.py`
Hyperparameter tuning takes too long	Too many combinations	Reduce `param_grid` in `models/base_model.py`
`Class imbalance` warning	Unequal class distribution	Add `class_weight='balanced'` to your estimator

🤝 Get Help & Connect

Building pipelines can be challenging - we're here to help!

Get architecture reviews, pipeline feedback, and join deep-dive sessions on ML engineering best practices. Share your pipeline adaptations and get inspired by others!

🎯 Ready for Your Next Step?

Starting classification?

→ Begin with Lecture_4.ipynb

Ready for pipelines?

→ Study the Pipeline/ example thoroughly

Want to test the complete system?

→ Run breast_cancer_pipline.py

Ready to build your own?

→ Create your project following the pipeline pattern

Ready to advance?

→ Continue to Module 3: Neural Networks from Scratch

📚 Reference Materials

Resource	Purpose	When to Use
`Pipeline/run_pipeline.py`	Complete pipeline example	Learning architecture
`Pipeline/streamlit_app/app.py`	Production deployment	Building your UI
`Pipeline/config/config.py`	Configuration template	Project setup
`breast_cancer_pipline.py`	Integrated example	Testing end-to-end flow

"السير" - "Walking on a road"
Professional ML is about systems, not just models. This pipeline example shows you the path from notebooks to production.

Study the pattern, then build your masterpiece! 🏗️

🔜 Next Step: Module 3: Neural Networks from Scratch

🗂️ Module Structure:

3_Classification/
│
├── 📚 README.md                          # This guide
├── 🎯 Lecture_4.ipynb                    # Classification from Scratch
├── 🚀 Lecture_5.ipynb                    # Production Pipeline Design
├── 🔧 breast_cancer_pipline.py           # Integrated Example
└── 🏗️ Pipeline/                         # Professional Architecture
    ├── config/                           # Configuration Management
    ├── data/                             # Data Processing
    ├── models/                           # ML Modeling
    ├── utils/                            # Shared Utilities
    ├── streamlit_app/                    # Deployment Interface
    ├── run_pipeline.py                   # Main Execution
    ├── README.md                         # Detailed Documentation
    └── requirements.txt                  # Dependencies

Uh oh!

FilesExpand file tree

README.md

Latest commit

History