Restructure configs: separate nodes/edges per scale (datasets schema constraint)

README.md

@@ -19,35 +19,46 @@ tags:
 - link-prediction
 - node-classification
 configs:
-- config_name: 1k
+- config_name: 1k_nodes
+  default: true
   data_files:
-  - split: nodes
+  - split: train
     path: 1k/nodes.parquet
-  - split: edges
+- config_name: 1k_edges
+  data_files:
+  - split: train
     path: 1k/edges.parquet
-- config_name: 10k
+- config_name: 10k_nodes
   data_files:
-  - split: nodes
+  - split: train
     path: 10k/nodes.parquet
-  - split: edges
+- config_name: 10k_edges
+  data_files:
+  - split: train
     path: 10k/edges.parquet
-- config_name: 100k
+- config_name: 100k_nodes
   data_files:
-  - split: nodes
+  - split: train
     path: 100k/nodes.parquet
-  - split: edges
+- config_name: 100k_edges
+  data_files:
+  - split: train
     path: 100k/edges.parquet
-- config_name: 1m
+- config_name: 1m_nodes
   data_files:
-  - split: nodes
+  - split: train
     path: 1m/nodes.parquet
-  - split: edges
+- config_name: 1m_edges
+  data_files:
+  - split: train
     path: 1m/edges.parquet
-- config_name: 10m
+- config_name: 10m_nodes
   data_files:
-  - split: nodes
+  - split: train
     path: 10m/nodes.parquet
-  - split: edges
+- config_name: 10m_edges
+  data_files:
+  - split: train
     path: 10m/edges.parquet
 ---
 
@@ -79,10 +90,12 @@ five separate graphs rather than nested cuts.
 
 ## Schema
 
-Each config has two splits, named `nodes` and `edges` (rather than the usual
-train/test/validation, since this is a graph dataset).
+Each scale exposes two configs, `<scale>_nodes` and `<scale>_edges`. They
+share a single split named `train` (a `datasets` convention — there is no
+held-out test split, since the intended use is to define your own splits over
+the graph).
 
-### `nodes` split
+### `nodes` config
 
 | Column       | Type   | Description                                                           |
 |--------------|--------|-----------------------------------------------------------------------|
@@ -98,7 +111,7 @@ The `attributes` JSON object has different keys depending on `node_type`:
 - **Venue**: `type` (string, e.g. `journal`, `conference`), `publisher` (string)
 - **Concept**: `domain` (string, e.g. `CS`)
 
-### `edges` split
+### `edges` config
 
 | Column     | Type   | Description                                                                                |
 |------------|--------|--------------------------------------------------------------------------------------------|
@@ -122,9 +135,10 @@ Relation semantics:
 ```python
 from datasets import load_dataset
 
-# Pick a scale: "1k", "10k", "100k", "1m", or "10m"
-nodes = load_dataset("jugalgajjar/CS-Knowledge-Graph-Dataset", "10k", split="nodes")
-edges = load_dataset("jugalgajjar/CS-Knowledge-Graph-Dataset", "10k", split="edges")
+# Configs follow the pattern "<scale>_nodes" / "<scale>_edges".
+# Scales: 1k, 10k, 100k, 1m, 10m
+nodes = load_dataset("jugalgajjar/CS-Knowledge-Graph-Dataset", "10k_nodes", split="train")
+edges = load_dataset("jugalgajjar/CS-Knowledge-Graph-Dataset", "10k_edges", split="train")
 
 print(nodes[0])
 # {'node_id': 'paper_W...', 'node_name': '...', 'node_type': 'Paper',
@@ -151,8 +165,8 @@ from torch_geometric.data import HeteroData
 from datasets import load_dataset
 
 scale = "10k"
-nodes = load_dataset("jugalgajjar/CS-Knowledge-Graph-Dataset", scale, split="nodes").to_pandas()
-edges = load_dataset("jugalgajjar/CS-Knowledge-Graph-Dataset", scale, split="edges").to_pandas()
+nodes = load_dataset("jugalgajjar/CS-Knowledge-Graph-Dataset", f"{scale}_nodes", split="train").to_pandas()
+edges = load_dataset("jugalgajjar/CS-Knowledge-Graph-Dataset", f"{scale}_edges", split="train").to_pandas()
 
 data = HeteroData()
 id_maps = {}