Datasets:

daqh
/

NDC-classes

@@ -12,6 +12,15 @@ task_categories:
 NDC-classes is an undirected hypergraph built from the U.S. FDA’s National Drug Code (NDC) Directory, designed for higher-order network / hypergraph machine learning in the drug domain. Each hyperedge corresponds to a drug and connects the set of pharmacologic/therapeutic class labels assigned to that drug, while nodes represent the class labels themselves (e.g., “serotonin reuptake inhibitor”), capturing co-classification patterns as higher-order interactions rather than pairwise links.
 <div align="center">
   <table>
     <tbody>
@@ -38,6 +47,8 @@ NDC-classes is an undirected hypergraph built from the U.S. FDA’s National Dru
   </table>
 </div>
 The hypergraph is stored in HIF (Hypergraph Interchange Format) as a JSON object, following the schema used to exchange higher-order network data across tools. Concretely, the dataset provides the canonical HIF fields-network-type, metadata, nodes, edges, and incidences-so you can reconstruct the full incidence structure without additional processing.
 In addition to the raw hypergraph topology, vector features are provided for both nodes and hyperedges (in their attribute dictionaries), enabling out-of-the-box experimentation with representation learning and downstream tasks:
@@ -45,7 +56,9 @@ In addition to the raw hypergraph topology, vector features are provided for bot
 - Node2Vec embeddings: random-walk–based structural embeddings.
 - VilLain embeddings: self-supervised hypergraph representation learning via virtual label propagation.
-Basic statistics (as packaged here): 1,161 nodes, 1,222 hyperedges, 183 connected component (Largest Connected Component Ratio 0.540913006029285).
 ```
 @article{Benson-2018-simplicial,

 NDC-classes is an undirected hypergraph built from the U.S. FDA’s National Drug Code (NDC) Directory, designed for higher-order network / hypergraph machine learning in the drug domain. Each hyperedge corresponds to a drug and connects the set of pharmacologic/therapeutic class labels assigned to that drug, while nodes represent the class labels themselves (e.g., “serotonin reuptake inhibitor”), capturing co-classification patterns as higher-order interactions rather than pairwise links.
+## Usage
+```python
+dataset = load_dataset("daqh/NDC-classes", split="full")
+hypergraphs = [xgi.from_hif_dict(d, nodetype=int, edgetype=int) for d in dataset]
+```
+## Statistics
 <div align="center">
   <table>
     <tbody>
   </table>
 </div>
+## Content
 The hypergraph is stored in HIF (Hypergraph Interchange Format) as a JSON object, following the schema used to exchange higher-order network data across tools. Concretely, the dataset provides the canonical HIF fields-network-type, metadata, nodes, edges, and incidences-so you can reconstruct the full incidence structure without additional processing.
 In addition to the raw hypergraph topology, vector features are provided for both nodes and hyperedges (in their attribute dictionaries), enabling out-of-the-box experimentation with representation learning and downstream tasks:
 - Node2Vec embeddings: random-walk–based structural embeddings.
 - VilLain embeddings: self-supervised hypergraph representation learning via virtual label propagation.
+Basic statistics (as packaged here): 327 nodes, 12,799 hyperedges, 1 connected component.
+## Citation
 ```
 @article{Benson-2018-simplicial,