Xinze-Li-Moqian commited on
Commit
7bb6da6
·
verified ·
1 Parent(s): 0531579

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +70 -90
README.md CHANGED
@@ -10,156 +10,136 @@ tags:
10
  - mathlib
11
  - dependency-graph
12
  - formal-verification
 
13
  size_categories:
14
  - 100K<n<1M
15
  ---
16
 
17
- # MathlibGraph: Mathlib Dependency Graph Dataset
18
 
19
- Dependency graph of [Mathlib](https://github.com/leanprover-community/mathlib4) (commit [`534cf0b`](https://github.com/leanprover-community/mathlib4/commit/534cf0b), 2 Feb 2026), the largest formal mathematics library for Lean 4. Lean version: `v4.28.0-rc1`.
 
 
 
 
 
 
 
 
 
 
 
 
 
20
 
21
  ## Files
22
 
23
- ### Core data (v1)
24
 
25
  | File | Rows | Description |
26
  |------|------|-------------|
27
- | `mathlib_nodes.csv` | 317,655 (308,129 unique) | Mathlib declarations (name, kind, module) |
28
- | `mathlib_edges.csv` | 8,436,366 | Mathlib dependencies (source, target, is_explicit, is_simplifier) |
29
- | `nodes.csv` | 633,364 | Full environment (Lean + Std + Mathlib) |
30
  | `edges.csv` | 10,889,011 | Full environment edges |
31
  | `mechanisms.ndjson` | — | Lean language mechanism extractions |
32
- | `tactic_usage.ndjson` | — | Tactic usage profiles |
33
 
34
- ### Enriched data (v2)
35
 
36
- | File | Rows | Description |
37
- |------|------|-------------|
38
- | `v2/mathlib_nodes_enriched.csv` | 308,129 | Deduplicated nodes with 8 precomputed metric columns |
39
- | `v2/mathlib_modules.csv` | 7,564 | Per-module (source file) metrics |
40
- | `v2/mathlib_namespaces_k2.csv` | 10,097 | Per-namespace (depth-2) metrics |
41
- | `v2/mathlib_summary.json` | — | Headline statistics for all three graph levels |
42
-
43
- > **Note:** `mathlib_nodes.csv` contains 9,526 duplicate name rows from `@[to_additive]` mirroring. The v2 enriched file is deduplicated to **308,129 unique declarations**.
44
-
45
- ## Statistics (308,129 unique Mathlib declarations)
46
-
47
- | Kind | Count | % |
48
- |------|-------|---|
49
- | theorem | 251,642 | 79.2% |
50
- | definition | 50,156 | 15.8% |
51
- | abbrev | 6,784 | 2.1% |
52
- | constructor | 4,755 | 1.5% |
53
- | inductive | 3,813 | 1.2% |
54
- | opaque | 499 | 0.2% |
55
- | quotient | 3 | <0.1% |
56
- | axiom | 3 | <0.1% |
57
 
58
  ## Quick Start
59
 
60
  ```python
61
  from datasets import load_dataset
62
 
63
- # Load enriched nodes (v2, deduplicated, with metrics)
64
  nodes = load_dataset("MathNetwork/MathlibGraph",
65
  data_files="v2/mathlib_nodes_enriched.csv", split="train").to_pandas()
66
  edges = load_dataset("MathNetwork/MathlibGraph",
67
  data_files="mathlib_edges.csv", split="train").to_pandas()
68
 
69
  print(f"Declarations: {len(nodes):,}, Edges: {len(edges):,}")
70
- # Declarations: 308,129, Edges: 8,436,366
71
-
72
- # Top 10 by PageRank
73
- print(nodes.nlargest(10, "pagerank")[["name", "kind", "in_degree", "pagerank"]])
74
-
75
- # Load module-level data
76
- modules = load_dataset("MathNetwork/MathlibGraph",
77
- data_files="v2/mathlib_modules.csv", split="train").to_pandas()
78
- print(f"Modules: {len(modules):,}")
79
  ```
80
 
81
  ## Schema
82
 
83
- ### mathlib_nodes.csv (v1)
84
-
85
- | Column | Type | Description |
86
- |--------|------|-------------|
87
- | name | string | Fully qualified declaration name |
88
- | kind | string | `theorem`, `definition`, `abbrev`, `inductive`, `constructor`, `opaque`, `axiom`, `quotient` |
89
- | module | string | Parent namespace of the declaration |
90
-
91
- ### mathlib_edges.csv
92
-
93
- | Column | Type | Description |
94
- |--------|------|-------------|
95
- | source | string | Dependent declaration |
96
- | target | string | Dependency (premise) |
97
- | is_explicit | bool | Appears in explicit arguments (25.8% of edges) |
98
- | is_simplifier | bool | Used by simplifier |
99
-
100
  ### v2/mathlib_nodes_enriched.csv
101
 
102
- All columns from `mathlib_nodes.csv` plus:
103
-
104
  | Column | Type | Description |
105
  |--------|------|-------------|
106
- | namespace_depth2 | string | Depth-2 namespace (e.g., `Mathlib.Algebra`) |
107
- | namespace_depth3 | string | Depth-3 namespace (e.g., `Mathlib.Algebra.Group`) |
108
- | in_degree | int | Number of declarations that depend on this one |
109
- | out_degree | int | Number of declarations this one depends on |
110
- | pagerank | float | PageRank score (alpha=0.85) |
 
 
 
111
  | betweenness | float | Betweenness centrality (sampled, k=500, seed=42) |
112
- | community_id | int | Louvain community ID (resolution=1.0, seed=42) |
113
- | dag_layer | int | Topological depth in DAG (-1 if in a cycle; 5,732 nodes) |
114
 
115
  ### v2/mathlib_modules.csv
116
 
117
- One row per Mathlib source file (7,564 modules).
118
-
119
  | Column | Type | Description |
120
  |--------|------|-------------|
121
- | module | string | Dotted module name (e.g., `Mathlib.Algebra.Group.Defs`) |
122
- | decl_count | int | Declarations in this module (from jixia extraction, 235K coverage) |
123
- | in_degree | int | Modules that import this module |
124
- | out_degree | int | Modules this module imports |
125
  | pagerank | float | PageRank on module import graph |
126
  | betweenness | float | Betweenness centrality (exact, no sampling) |
127
  | dag_layer | int | Topological depth in module DAG |
128
- | community_id | int | Louvain community ID |
129
- | cohesion | float | internal_edges / (internal + external) |
130
  | import_utilization_median | float | Median fraction of imported declarations actually used |
131
 
132
  ### v2/mathlib_namespaces_k2.csv
133
 
134
- One row per depth-2 namespace (10,097 namespaces).
135
-
136
  | Column | Type | Description |
137
  |--------|------|-------------|
138
- | namespace | string | Depth-2 namespace (e.g., `Mathlib.Algebra`) |
139
  | decl_count | int | Declarations in this namespace |
140
  | in_degree | int | Unweighted in-degree in namespace graph |
141
- | out_degree | int | Unweighted out-degree in namespace graph |
142
- | edge_weight_sum | int | Weighted degree (sum of declaration-level edges) |
143
- | pagerank | float | PageRank (weighted, alpha=0.85) |
144
- | betweenness | float | Betweenness centrality (k=300, seed=42, weighted) |
145
- | community_id | int | Louvain community ID (weighted undirected) |
146
- | cross_ns_ratio | float | Fraction of this namespace's edges that cross namespace boundaries |
147
 
148
  ### v2/mathlib_summary.json
149
 
150
- Headline statistics including: snapshot metadata (commit, date, Lean version), declaration/module/namespace graph metrics (node/edge counts, modularity, NMI, DAG depth, power-law exponents), and edge decomposition ratios.
151
 
152
- ## Extraction Pipeline
153
 
154
- - **Nodes**: [lean4export](https://github.com/leanprover/lean4export)
155
- - **Edges**: [lean-training-data premises](https://github.com/kim-em/lean-training-data)
156
- - **Module graph**: [importGraph](https://github.com/leanprover-community/importGraph)
157
- - **Enrichment**: NetworkX (PageRank, betweenness, Louvain, DAG layers)
158
- - **Code**: [MathlibGraph on GitHub](https://github.com/MathNetwork/MathlibGraph)
 
 
159
 
160
  ## Citation
161
 
162
- Part of the [MathlibGraph](https://github.com/MathNetwork/MathlibGraph) project — network analysis of formalized mathematics.
 
 
 
 
 
 
 
 
163
 
164
  ## License
165
 
 
10
  - mathlib
11
  - dependency-graph
12
  - formal-verification
13
+ - network-analysis
14
  size_categories:
15
  - 100K<n<1M
16
  ---
17
 
18
+ # MathlibGraph: The Multinetwork of Mathlib
19
 
20
+ Dependency graph of [Mathlib](https://github.com/leanprover-community/mathlib4) (commit [`534cf0b`](https://github.com/leanprover-community/mathlib4/commit/534cf0b), 2 Feb 2026), the largest formal mathematics library for Lean 4 (`v4.28.0-rc1`).
21
+
22
+ Three dependency layers (declarations, modules, namespaces) with precomputed network metrics.
23
+
24
+ ## Quick Stats
25
+
26
+ | | Declarations | Modules | Namespaces (k=2) |
27
+ |---|---|---|---|
28
+ | Nodes | 308,129 | 7,564 | 10,097 |
29
+ | Edges | 8,436,366 | 20,881 | 332,081 |
30
+ | DAG depth | 83 | 154 | 7 |
31
+ | Louvain modularity | 0.478 | 0.610 | 0.270 |
32
+ | Synthesized edges | 74.2% | — | — |
33
+ | Cross-namespace edges | — | — | 85.8% |
34
 
35
  ## Files
36
 
37
+ ### Raw Graph Data
38
 
39
  | File | Rows | Description |
40
  |------|------|-------------|
41
+ | `mathlib_nodes.csv` | 317,655 | Declarations before deduplication (name, kind, module) |
42
+ | `mathlib_edges.csv` | 8,436,366 | Declaration dependencies (source, target, is_explicit, is_simplifier) |
43
+ | `nodes.csv` | 633,364 | Full environment including Lean core and Std |
44
  | `edges.csv` | 10,889,011 | Full environment edges |
45
  | `mechanisms.ndjson` | — | Lean language mechanism extractions |
46
+ | `tactic_usage.ndjson` | — | Per-declaration tactic usage profiles |
47
 
48
+ ### Enriched Data (v2)
49
 
50
+ | File | Rows | Columns | Description |
51
+ |------|------|---------|-------------|
52
+ | `v2/mathlib_nodes_enriched.csv` | 308,129 | 11 | Deduplicated declarations with network metrics |
53
+ | `v2/mathlib_modules.csv` | 7,564 | 10 | Per-module (source file) metrics |
54
+ | `v2/mathlib_namespaces_k2.csv` | 10,097 | 9 | Per-namespace (depth-2) metrics |
55
+ | `v2/mathlib_summary.json` | — | — | Headline statistics for all three graph levels |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
56
 
57
  ## Quick Start
58
 
59
  ```python
60
  from datasets import load_dataset
61
 
 
62
  nodes = load_dataset("MathNetwork/MathlibGraph",
63
  data_files="v2/mathlib_nodes_enriched.csv", split="train").to_pandas()
64
  edges = load_dataset("MathNetwork/MathlibGraph",
65
  data_files="mathlib_edges.csv", split="train").to_pandas()
66
 
67
  print(f"Declarations: {len(nodes):,}, Edges: {len(edges):,}")
68
+ print(nodes.nlargest(5, "pagerank")[["name", "kind", "in_degree", "pagerank"]])
 
 
 
 
 
 
 
 
69
  ```
70
 
71
  ## Schema
72
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
73
  ### v2/mathlib_nodes_enriched.csv
74
 
 
 
75
  | Column | Type | Description |
76
  |--------|------|-------------|
77
+ | name | str | Fully qualified declaration name |
78
+ | kind | str | One of: theorem, definition, abbrev, inductive, constructor, opaque, axiom, quotient |
79
+ | module | str | Parent namespace of the declaration (null for 30,944 Lean core declarations) |
80
+ | namespace_depth2 | str | First 2 dot-separated name components (e.g., `Mathlib.Algebra`) |
81
+ | namespace_depth3 | str | First 3 dot-separated name components |
82
+ | in_degree | int | Declarations that depend on this one |
83
+ | out_degree | int | Declarations this one depends on |
84
+ | pagerank | float | PageRank (alpha=0.85) |
85
  | betweenness | float | Betweenness centrality (sampled, k=500, seed=42) |
86
+ | community_id | int | Louvain community assignment |
87
+ | dag_layer | int | Topological depth; -1 for 5,732 nodes in cycles |
88
 
89
  ### v2/mathlib_modules.csv
90
 
 
 
91
  | Column | Type | Description |
92
  |--------|------|-------------|
93
+ | module | str | Dotted module name (e.g., `Mathlib.Algebra.Group.Defs`) |
94
+ | decl_count | int | Declarations in this module (235K jixia coverage) |
95
+ | in_degree | int | Modules that import this one |
96
+ | out_degree | int | Modules this one imports |
97
  | pagerank | float | PageRank on module import graph |
98
  | betweenness | float | Betweenness centrality (exact, no sampling) |
99
  | dag_layer | int | Topological depth in module DAG |
100
+ | community_id | int | Louvain community assignment |
101
+ | cohesion | float | Fraction of edges that stay within the module |
102
  | import_utilization_median | float | Median fraction of imported declarations actually used |
103
 
104
  ### v2/mathlib_namespaces_k2.csv
105
 
 
 
106
  | Column | Type | Description |
107
  |--------|------|-------------|
108
+ | namespace | str | Depth-2 namespace (e.g., `Mathlib.Algebra`) |
109
  | decl_count | int | Declarations in this namespace |
110
  | in_degree | int | Unweighted in-degree in namespace graph |
111
+ | out_degree | int | Unweighted out-degree |
112
+ | edge_weight_sum | int | Sum of declaration-level edges involving this namespace |
113
+ | pagerank | float | Weighted PageRank (alpha=0.85) |
114
+ | betweenness | float | Weighted betweenness (k=300, seed=42) |
115
+ | community_id | int | Louvain community on weighted undirected graph |
116
+ | cross_ns_ratio | float | Fraction of edges crossing namespace boundaries |
117
 
118
  ### v2/mathlib_summary.json
119
 
120
+ Top-level keys: `snapshot` (commit, date, Lean version), `declaration_graph`, `module_graph`, `namespace_graph_k2` (node/edge counts, modularity, NMI, DAG depth, power-law exponents), `edge_decompositions` (proof-only/statement-only ratios), `kind_distribution`.
121
 
122
+ ## Methodology
123
 
124
+ - **Extraction**: lean4export (nodes) + lean-training-data (edges) + importGraph (module graph) + jixia (declaration metadata)
125
+ - **Deduplication**: by name, 317,655 to 308,129 rows (9,526 `@[to_additive]` mirrors removed)
126
+ - **Self-loops**: 4,755 constructor self-references filtered from graph construction
127
+ - **PageRank**: alpha=0.85, max_iter=100, tol=1e-6
128
+ - **Betweenness**: declaration k=500, namespace k=300, module exact; all with seed=42
129
+ - **Communities**: Louvain (python-louvain), resolution=1.0, random_state=42, on undirected projection
130
+ - **DAG layers**: Kahn's algorithm; 5,732 declaration nodes in cycles assigned layer=-1; namespace graph condensed via SCC
131
 
132
  ## Citation
133
 
134
+ ```bibtex
135
+ @inproceedings{mathlibgraph2026,
136
+ title = {The Multinetwork of Mathlib},
137
+ author = {Li, Xinze and Shafto, Patrick},
138
+ booktitle = {NeurIPS 2026 Datasets and Benchmarks Track},
139
+ year = {2026},
140
+ note = {Under review}
141
+ }
142
+ ```
143
 
144
  ## License
145