TimeGPT, Tabula, and RelBench Project

This repository contains Colab notebooks showcasing machine learning tasks using TimeGPT, Tabula, and RelBench. Each notebook demonstrates various capabilities, including forecasting, anomaly detection, synthetic data generation, and tabular prediction. A video presentation walks through each notebook’s purpose, code, and outputs.

Table of Contents

• TimeGPT
  • Multivariate & Long Horizon Forecasting
  • Fine-tuning with Custom Data
  • Anomaly Detection
  • Energy Forecasting
  • Bitcoin Price Prediction
• Tabular
  • Synthetic Data Generation
  • Zero-shot Inference
• RDL and RelBench
  • GNN-based Model Training for Tabular Prediction

---

TimeGPT

Multivariate & Long Horizon Forecasting

• Notebook: Multivariate_long_horizon.ipynb
• Description: Forecast multiple time series over long horizons to understand relationships across variables.
• Steps: Data preparation, model configuration, forecasting, and results visualization.

Fine-tuning with Custom Data

• Notebook: finetune.ipynb
• Description: Fine-tune TimeGPT for specific time series tasks to enhance accuracy.
• Steps: Load custom data, run fine-tuning, and evaluate predictions.

Anomaly Detection

• Notebook: Anomaly_detection.ipynb
• Description: Detect anomalies in time series data to identify significant deviations.
• Steps: Dataset preparation, configure detection, and visualize anomalies.

Energy Forecasting

• Notebook: Energy_forecasting.ipynb
• Description: Forecast energy demand to manage resources efficiently.
• Steps: Load energy data, configure forecasting model, and analyze results.

Bitcoin Price Prediction

• Notebook: Bitcoin_forecasting.ipynb
• Description: Predict Bitcoin prices, demonstrating TimeGPT’s capability in volatile data handling.
• Steps: Import Bitcoin data, train the model, and visualize forecasts.

Tabular

Synthetic Data Generation

• Notebook: synthetic_data_for_a_real_data_set.ipynb
• Description: Generate synthetic data that resembles real datasets to aid in model training.
• Steps: Load real data, create synthetic dataset, and compare for feature consistency.

Zero-shot Inference

• Notebook: zero_shot_inference.ipynb
• Description: Perform zero-shot inference, showcasing Tabula’s predictive capability without prior dataset training.
• Steps: Model setup, data input, and analyze model generalization.

RDL and RelBench

GNN-based Model Training for Tabular Prediction

• Notebook: train_model.ipynb
• Description: Train a GNN model using RelBench for relational data predictions.
• Steps: Prepare tabular data, train GNN, and evaluate performance metrics.

---

🚀 Data Mining with AutoGluon

Explore various AutoGluon implementations on real-world datasets. Below is a collection of interactive notebooks that demonstrate various machine learning techniques, from tabular classification to time series forecasting. Click on the links to explore the live notebooks!

🎥 Video Walkthrough:

Watch the YouTube Tutorial

📊 Assignment Notebooks

Task	Notebook Link
Autogluon with Kaggle dataset	Link
Tabular Classification	Link
Multimodal Tabular	Link
Multilabel Prediction	Link
Training Models with GPU Support	Link
Segment Analysis & Sentence Similarity	Link
Finetune Foundation Models	Link
Named Entity Recognition (NER)	Link
Clip Zero Shot	Link
Image Object Detection	Link
Text-to-Text Semantic Matching	Link
Multimodal Mixed Text Columns	Link
Multimodal Images & Text	Link
Forecasting Time Series - In Depth	Link
Document Classification	Link
PDF Classification	Link
Time Series Forecasting	Link