dttutty commited on
Commit
9ccaf18
·
1 Parent(s): c3aa290

docs: generate dataset catalog from manifests

Browse files
Files changed (2) hide show
  1. DATASET_CATALOG.md +20 -0
  2. README.md +13 -44
DATASET_CATALOG.md ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Dataset Catalog
2
+
3
+ This file is generated from each dataset's `manifest.json` by FROST.
4
+ Do not edit the table by hand.
5
+
6
+ | Dataset | Edges | Train | Validation | Test | Max eval macro-batch | Sampling adjacency | Optional artifacts |
7
+ | --- | ---: | ---: | ---: | ---: | ---: | --- | --- |
8
+ | CANPARL | 74,478 | 52,556 | 11,809 | 10,113 | 10,113 | bidirectional | `edge_features.npy` (74479x1, int16) |
9
+ | CONTACTS | 2,426,279 | 1,698,368 | 363,881 | 364,030 | 363,881 | bidirectional | `edge_features.npy` (2426280x1, int8) |
10
+ | ENRON | 125,235 | 87,664 | 18,786 | 18,785 | 18,785 | bidirectional | — |
11
+ | FLIGHTS | 1,927,145 | 1,343,095 | 296,226 | 287,824 | 287,824 | bidirectional | `edge_features.npy` (1927146x1, int16) |
12
+ | LASTFM | 1,293,103 | 905,172 | 193,966 | 193,965 | 193,965 | bidirectional | `node_role.npy` (1981x1, bool) |
13
+ | MOOC | 411,749 | 288,224 | 61,763 | 61,762 | 61,762 | bidirectional | `edge_features.npy` (411750x4, float64)<br>`node_role.npy` (7145x1, bool) |
14
+ | REDDIT | 672,447 | 470,713 | 100,867 | 100,867 | 100,867 | bidirectional | `edge_features.npy` (672448x172, float64)<br>`node_role.npy` (10985x1, bool) |
15
+ | SOCIALEVO | 2,099,519 | 1,469,665 | 314,924 | 314,930 | 314,924 | bidirectional | `edge_features.npy` (2099520x2, float64) |
16
+ | UCI | 59,835 | 41,884 | 8,976 | 8,975 | 8,975 | bidirectional | — |
17
+ | UNTRADE | 507,497 | 348,431 | 77,079 | 81,987 | 77,079 | bidirectional | `edge_features.npy` (507498x1, int32) |
18
+ | UNVOTE | 1,035,742 | 722,633 | 157,990 | 155,119 | 155,119 | bidirectional | `edge_features.npy` (1035743x1, int16) |
19
+ | USLEGIS | 60,396 | 40,491 | 10,101 | 9,804 | 9,804 | bidirectional | `edge_features.npy` (60397x1, int16) |
20
+ | WIKIPEDIA | 157,474 | 110,232 | 23,621 | 23,621 | 23,621 | bidirectional | `edge_features.npy` (157475x172, float64)<br>`node_role.npy` (9228x1, bool) |
README.md CHANGED
@@ -127,47 +127,16 @@ If you need a `datasets` library dataset or a Hub Viewer-backed table, export th
127
  - Selected integer-valued edge-feature arrays are materialized with lossless downcasts for storage efficiency.
128
  - Fixed negatives are evaluation-only destination corruptions. Training uses a fresh random destination sample and should not reuse the fixed file every epoch.
129
 
130
- ## Recommended `max_macro_batch_size`
131
-
132
- These conservative recommendations use the smaller actual validation/test
133
- split after timestamp groups have been kept intact. Exact counts are recorded
134
- in each dataset's `manifest.json`.
135
-
136
- | Dataset | max_macro_batch_size |
137
- | --- | ---: |
138
- | CanParl | 10113 |
139
- | Contacts | 363881 |
140
- | Flights | 287824 |
141
- | SocialEvo | 314924 |
142
- | UNtrade | 77079 |
143
- | UNvote | 155119 |
144
- | USLegis | 9804 |
145
- | Enron | 18785 |
146
- | LastFM | 193965 |
147
- | MOOC | 61762 |
148
- | Reddit | 100867 |
149
- | UCI | 8975 |
150
- | Wikipedia | 23621 |
151
-
152
- ## Dataset Details
153
-
154
- - Source or destination ranges are computed from `*/edges.csv` (`src`, `dst`).
155
- - The curated bundle already reflects timestamp normalization for `CanParl`, `UNtrade`, and `UNvote` and downcast integer edge-feature materialization for `CanParl`, `Contacts`, `Flights`, `UNtrade`, `UNvote`, and `USLegis`.
156
- - DGB paper: [https://arxiv.org/pdf/2207.10128](https://arxiv.org/pdf/2207.10128)
157
- - `Optional Artifacts` lists only non-`edges.csv` sidecar artifacts, shown as `size, rowsxcols, dtype`.
158
-
159
- | Dataset | SRC_NID | DST_NID | Notes | Optional Artifacts | TS INFO |
160
- | --- | --- | --- | --- | --- | --- |
161
- | [CanParl](CANPARL/edges.csv) (num_edges: 74,478) | range: 1->734<br>unique: 734 | range: 2->734<br>unique: 244 | Canadian MP interaction network.<br>Edge weight = yearly count of shared "yes" votes on bills. | [edge_features.npy](CANPARL/edge_features.npy) 145.6KB, 74,479x1, int16 | yearly<br>range=[0->13]<br>unique=14 |
162
- | [Contacts](CONTACTS/edges.csv) (num_edges: 2,426,279) | range: 1->692<br>unique: 676 | range: 1->690<br>unique: 676 | University-student physical proximity network over one month.<br>Edge weight = proximity strength. | [edge_features.npy](CONTACTS/edge_features.npy) 2.3MB, 2,426,280x1, int8 | Second<br>range=[0->2418900]<br>unique=8064 |
163
- | [Flights](FLIGHTS/edges.csv) (num_edges: 1,927,145) | range: 1->13169<br>unique: 11574 | range: 1->13169<br>unique: 12939 | Airport traffic during COVID-19.<br>Edge weight = number of flights between two airports in a day. | [edge_features.npy](FLIGHTS/edge_features.npy) 3.7MB, 1,927,146x1, int16 | daily<br>range=[0->121]<br>unique=122 |
164
- | [SocialEvo](SOCIALEVO/edges.csv) (num_edges: 2,099,519) | range: 1->74<br>unique: 74 | range: 1->74<br>unique: 70 | Mobile phone proximity network in an undergraduate dorm over eight months.<br>Each edge has a 2-dim feature. | [edge_features.npy](SOCIALEVO/edge_features.npy) 32.0MB, 2,099,520x2, float64 | Second<br>range=[0->20,935,623]<br>unique=565,932 |
165
- | [UNtrade](UNTRADE/edges.csv) (num_edges: 507,497) | range: 1->255<br>unique: 255 | range: 1->255<br>unique: 254 | Food and agriculture trade between nations over 30+ years.<br>Edge weight = normalized import or export value. | [edge_features.npy](UNTRADE/edge_features.npy) 1.9MB, 507,498x1, int32 | yearly<br>range=[0->31]<br>unique=32 |
166
- | [UNvote](UNVOTE/edges.csv) (num_edges: 1,035,742) | range: 1->201<br>unique: 201 | range: 1->201<br>unique: 201 | UN General Assembly roll-call votes.<br>Edge weight increases when two nations both vote "yes". | [edge_features.npy](UNVOTE/edge_features.npy) 2.0MB, 1,035,743x1, int16 | yearly<br>range=[0->71]<br>unique=72 |
167
- | [USLegis](USLEGIS/edges.csv) (num_edges: 60,396) | range: 1->225<br>unique: 224 | range: 1->225<br>unique: 225 | US Senate co-sponsorship network.<br>Edge weight = number of shared bill co-sponsorships in a congress. | [edge_features.npy](USLEGIS/edge_features.npy) 118.1KB, 60,397x1, int16 | bi-yearly<br>range=[0->11]<br>unique=12 |
168
- | [Enron](ENRON/edges.csv) (num_edges: 125,235) | range: 1->184<br>unique: 181 | range: 1->184<br>unique: 184 | Email communications between Enron employees over three years. | n/a | Second<br>range=[0->113,740,399]<br>unique=22,632 |
169
- | [LastFM](LASTFM/edges.csv) (num_edges: 1,293,103) | range: 1->980<br>unique: 980 | range: 981->1980<br>unique: 1000 | Bipartite user-song listening graph over one month. | [node_role.npy](LASTFM/node_role.npy) 2.1KB, 1,981x1, bool | Second<br>range=[0->137,107,267]<br>unique=1,283,614 |
170
- | [MOOC](MOOC/edges.csv) (num_edges: 411,749) | range: 1->7047<br>unique: 7047 | range: 7048->7144<br>unique: 97 | Bipartite student-content interaction graph.<br>Each edge has a 4-dim feature. | [edge_features.npy](MOOC/edge_features.npy) 12.6MB, 411,750x4, float64<br>[node_role.npy](MOOC/node_role.npy) 7.1KB, 7,145x1, bool | Second<br>range=[0->2,572,086]<br>unique=345,600 |
171
- | [Reddit](REDDIT/edges.csv) (num_edges: 672,447) | range: 1->10000<br>unique: 10000 | range: 10001->10984<br>unique: 984 | Bipartite user-subreddit posting graph over one month.<br>172-dim LIWC edge feature. | [edge_features.npy](REDDIT/edge_features.npy) 882.4MB, 672,448x172, float64<br>[node_role.npy](REDDIT/node_role.npy) 10.9KB, 10,985x1, bool | Millisecond<br>range=[0->2,678,390,016]<br>unique=669,065 |
172
- | [UCI](UCI/edges.csv) (num_edges: 59,835) | range: 1->1899<br>unique: 1350 | range: 1->1898<br>unique: 1862 | Online communication network where nodes are university students.<br>Edges are posted messages. | n/a | Second<br>range=[0->16,736,181]<br>unique=58,911 |
173
- | [Wikipedia](WIKIPEDIA/edges.csv) (num_edges: 157,474) | range: 1->8227<br>unique: 8227 | range: 8228->9227<br>unique: 1000 | Bipartite user-page editing graph over one month.<br>172-dim LIWC edge feature. | [edge_features.npy](WIKIPEDIA/edge_features.npy) 206.6MB, 157,475x172, float64<br>[node_role.npy](WIKIPEDIA/node_role.npy) 9.1KB, 9,228x1, bool | Second<br>range=[0->2,678,373]<br>unique=152,757 |
 
127
  - Selected integer-valued edge-feature arrays are materialized with lossless downcasts for storage efficiency.
128
  - Fixed negatives are evaluation-only destination corruptions. Training uses a fresh random destination sample and should not reuse the fixed file every epoch.
129
 
130
+ ## Dataset Catalog
131
+
132
+ Exact edge counts, split sizes, evaluation batch ceilings, graph direction, and
133
+ artifact shapes are generated from the per-dataset manifests in
134
+ [`DATASET_CATALOG.md`](DATASET_CATALOG.md). Regenerate it from a FROST checkout:
135
+
136
+ ```bash
137
+ uv run frost-data catalog /path/to/frost_dataset \
138
+ --markdown-output /path/to/frost_dataset/DATASET_CATALOG.md --force
139
+ ```
140
+
141
+ Training defaults remain in FROST's `benchmarks/specs/presets.yaml`; the
142
+ generated maximum is only the smaller actual validation/test split.