UniKT

A unified knowledge tracing experiment framework — 35 built-in knowledge tracing models (GNN, attention, Mamba, etc.), covering the full workflow of training, evaluation, hyperparameter search, and case analysis.

Quick navigation

• Quick start — Installation and your first training run

• Model zoo — Performance comparison of models on standard datasets

• Framework data flow — End-to-end architecture walkthrough

Multi-model support

• 35 built-in knowledge tracing models (GNN, attention, Mamba)

• Decorator registration + static discovery with lazy loading

• Unified trainer API

Experiment management

• K-fold cross validation

• Optuna-based hyperparameter search

• SwanLab experiment tracking

Data processing

• 11 supported datasets

• Automatic download and preprocessing

• Configurable sampling strategies

• Standardized Parquet output

Case analysis

• Model inference and prediction

• Strategy-based student selection

• Knowledge state heatmaps