Add task category, paper/code links, dataset description, and data generation example

This PR significantly improves the dataset card for `shengyuf/RLD4CO` by:
- Adding the `graph-ml` task category and relevant tags (`combinatorial-optimization`, `graph-generation`) for better discoverability on the Hugging Face Hub.
- Including a direct link to the paper ([https://huggingface.co/papers/2502.00277](https://huggingface.co/papers/2502.00277)) for comprehensive research context.
- Linking to the associated GitHub repository ([https://github.com/Shengyu-Feng/RLD4CO](https://github.com/Shengyu-Feng/RLD4CO)) for access to the code and further implementation details.
- Providing a descriptive introduction to the dataset's contents, clarifying that it provides test data and data generation scripts for combinatorial optimization problems.
- Incorporating a sample usage snippet for data generation, as explicitly found in the GitHub README, to guide users on how to generate similar graph instances.
- Adding the official BibTeX citation for proper academic attribution.

README.md

@@ -1,3 +1,51 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+task_categories:
+- graph-ml
+tags:
+- combinatorial-optimization
+- graph-generation
+---
+
+This repository contains the test data and data generation scripts for combinatorial optimization problems presented in the paper [Regularized Langevin Dynamics for Combinatorial Optimization](https://huggingface.co/papers/2502.00277).
+
+The paper proposes Regularized Langevin Dynamics (RLD) as an effective sampling framework for combinatorial optimization, developing two solvers: RLSA (simulated annealing) and RLNN (neural network based). This dataset provides the graph instances used for evaluating these methods.
+
+Code: [https://github.com/Shengyu-Feng/RLD4CO](https://github.com/Shengyu-Feng/RLD4CO)
+
+### Data Overview
+
+We provide random graph generators in the associated GitHub repository for three classic combinatorial optimization problems, yielding various graph types and sizes:
+
+-   **Erdős–Rényi (ER):** ER-[700-800] (for training/validation/testing) and ER-[9000-11000] (for out-of-distribution testing). The ER test graphs are sourced from the [DIMES](https://github.com/DIMESTeam/DIMES) repository.
+-   **Revised Model B (RB):** RB-[200-300] and RB-[800-1200] (for training/validation/testing).
+-   **Barabási–Albert (BA):** BA-[200-300] and BA-[800-1200] (for training/validation/testing).
+
+The RB and BA graphs are generated using code from [DIffUCO](https://github.com/ml-jku/DIffUCO). For fair comparison, the test data comprising 1000 instances per problem (split into two groups of 500-size test sets) can be directly downloaded from this repository.
+
+### Data Generation Sample Usage
+
+You can generate similar graph instances using the provided scripts in the `data` folder of the GitHub repository. For example, to generate Erdős–Rényi graphs:
+
+```bash
+cd data
+
+python er_generator.py \
+  --num-graphs 1000 \
+  --min-n 700 --max-n 800 \
+  --seed 0 \
+  --save-dir ./ER_small_train \
+  --processes 10
+```
+
+### Citation
+
+```bibtex
+@inproceedings{feng2025regularizedlangevindynamicscombinatorial,
+      title={Regularized Langevin Dynamics for Combinatorial Optimization}, 
+      author={Shengyu Feng and Yiming Yang},
+      booktitle={International conference on machine learning},
+      year={2025},
+      organization={PMLR}
+}
+```