hsafaai/vdc-denoiser-m64-v1

VDC pretrained denoiser m64 v1

This repository contains the official pretrained checkpoint for the VDC paper model.

Model id: vdc-denoiser-m64-v1 Suggested Hugging Face repo id: hsafaai/vdc-denoiser-m64-v1

What This Is

• Method tag: denoiser_cond_enhanced
• Model type: denoiser
• Checkpoint step: 190000
• Selection rule: joint_best_standard_plus_complex
• Frozen on: 2026-02-13
• SHA256: 8a1b71ad6d2012ed344b22a743a96e667d32a2ab54ade4444686e375e91f5c0f

Files

• manifest.json: portable manifest for the released checkpoint
• train_config.yaml: exact training config embedded in the paper run
• model_selection_joint_best.json: checkpoint-selection provenance
• vdc-denoiser-m64-v1.pt: pretrained weights

Intended Use

• bivariate copula density estimation
• vine copula fitting with a pretrained pair-copula estimator
• mutual information and total correlation estimation

Limitations

• continuous marginals
• simplifying assumption for vine copulas
• trained on synthetic copula zoo rather than raw mixed-type tabular data

Usage

python scripts/download_pretrained.py --model-id vdc-denoiser-m64-v1 --repo-id hsafaai/vdc-denoiser-m64-v1
python examples/use_pretrained_model.py --model-id vdc-denoiser-m64-v1 --repo-id hsafaai/vdc-denoiser-m64-v1