AutoGluon Masterclass: Applied Machine Learning & Deep Learning

An end-to-end practical showcase of automated machine learning (AutoML) using AutoGluon—Amazon's state-of-the-art library for training highly accurate machine learning models with just a few lines of code.

This repository contains a structured, modular set of Jupyter notebooks covering real-world machine learning paradigms: Tabular Data, Multimodal Fusion (Text + Images + Tabular), Time-Series Forecasting, and Computer Vision.

---

🚀 Key Modules & Project Architecture

The repository is organized into four core directories, each representing a distinct AutoML application domain:

1. 📊 Tabular Data (00-Tabular-Data)

Advanced tabular modeling using automated feature engineering, ensemble selection, and multi-layer stacking.

• Key Tasks: Binary classification, multi-class classification, regression, and hyperparameter tuning presets.
• Datasets: Uber rides fare regression and hotel reservation cancellations.
• Notebooks:
  • 01-Classification.ipynb - AutoGluon Tabular classification fundamentals.
  • 02-Regression.ipynb - Building optimized regression architectures.
  • 03-Advanced.ipynb - Model ensembling, custom feature engineering, and custom metrics.

🧩 2. Multimodal Fusion (01-Multi-Modal)

Fusing heterogeneous data sources (text, raw images, and tabular columns) into unified deep learning backends.

• Key Tasks: Natural Language Processing (NLP) classification, Multimodal fusion, and custom embedding extraction.
• Datasets: PetFinder adoption prediction, Kindle reviews, and text sentiment analysis.
• Notebooks:
  • 00-Multi-Modal-Natural Language for MultiClass.ipynb
  • 01-Multi-Modal-Natural-Language-Binary-Classification.ipynb
  • 02-MultiModal-Image-Text-and-Tabular.ipynb - Unified training on a mixture of tabular properties, textual descriptions, and images.

📈 3. Time Series Forecasting (02-Time-Series-Forecasting)

Univariate and multivariate forecasting using deep learning models (DeepAR, Temporal Fusion Transformers, etc.).

• Key Tasks: Time-series forecasting incorporating past covariates and known future covariates.
• Datasets: Air quality metrics, alcohol sales trends, and restaurant visitor traffic.
• Notebooks:
  • 00-Single-Variate-Time-Series-Forecasting.ipynb - Univariate baseline forecasting.
  • 01-Known-Covariates-MultiVariate-Time-Series-Forecasting.ipynb - Integrating known future covariates (e.g. holidays, weather predictions).
  • 02-Past-Covariates-MultiVariate-Time-Series-Forecasting .ipynb - Incorporating historical past variables to boost forecast accuracy.

🖼️ 4. Computer Vision & Embeddings (03-Images)

Deep Learning models applied to image classification and utilizing pre-trained foundation models.

• Key Tasks: Image classification and zero-shot similarity using OpenAI's CLIP model.
• Notebooks:
  • 00-Image-Classification.ipynb - Fine-tuning vision models on custom datasets.
  • 01-OpenAI-CLIP-with-Autogluon.ipynb - Leveraging OpenAI CLIP to extract embeddings and perform zero-shot classification.

---

🛠️ Installation & Setup

This project uses modern python package management via uv or pip.

Prerequisites

• Python 3.9, 3.10, or 3.11
• CUDA-compatible GPU (Highly recommended for Multimodal and Computer Vision tasks)

Step-by-Step Run Guide

1. Clone the repository:

   git clone https://github.com/c-ibarra/autoGluon-Course.git
   cd autoGluon-Course

2. Create and activate a virtual environment: Using uv (Recommended):

   uv venv
   source .venv/bin/activate

   Or standard venv:

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Launch Jupyter:

   jupyter notebook

---

🧪 AutoML Models Utilized

AutoGluon automatically tests and ensembles a diverse range of architectures:

• Tabular: LightGBM, XGBoost, CatBoost, Random Forests, Extra Trees, and custom PyTorch Neural Networks.
• Text/Multimodal: Transformer backends (DeBERTa, ELECTRA) paired with tabular feature channels.
• Vision: ResNet, EfficientNet, ConvNeXt, and ViT (Vision Transformers).
• Time Series: AutoARIMA, Prophet, DeepAR, TFT (Temporal Fusion Transformer), and PatchTST.

---

💼 Portfolio Integration

This project is featured as part of my Applied Machine Learning Portfolio. For more projects, case studies, and to get in touch, please visit my homepage:

👉 c-ibarra Portfolio Homepage

---