Library Components

Overview

The library components module provides specialized implementations and utilities for various deep learning models in TALENT. These components include attention mechanisms, feature transformers, model-specific optimizations, and data processing utilities.

Common Features

All library components in TALENT share the following features:

  • PyTorch integration for deep learning models

  • Efficient data processing and memory optimization

  • Modular design for easy integration

  • Support for both numerical and categorical features

  • Configurable hyperparameters and architectures

  • GPU acceleration support

Available Components

Core Utilities:

  • Data: Functions for loading, preprocessing, and preparing tabular data for machine learning tasks, including handling missing values, encoding features, and creating data loaders.

  • TData: Optimized data structure for efficient tabular data handling

  • num_embeddings: Advanced numerical feature embedding techniques

Library Components:

  • TabNet: Interpretable deep learning for tabular data

  • TabPFN: Prior-data fitted networks

  • TabR: Tabular representation learning

  • TabM: Tabular modeling with transformers

  • RealMLP: Real-valued MLP for tabular data

  • BiSHop: Bidirectional hierarchical attention

  • NODE: Neural oblivious decision ensembles

  • HyperFast: Fast hyperparameter optimization

  • ExcelFormer: Transformer for tabular data

  • DANets: Deep attention networks

  • TabCaps: Capsule networks for tabular data

  • TabICL: In-context learning for tabular data

  • Periodic Tabular DL: Periodic embeddings for tabular data

  • TROMPT: Tabular prompting mechanisms

  • PTARL: Policy gradient methods for tabular RL

  • AmFormer: Attention mechanisms for transformers

  • TabPTM: Pre-trained models for tabular data

  • DNNR: Deep nearest neighbor regression

Library Components: