Extend dataset summary to create stats for each node/edge type

[kjkozlowski]

example output for hetero data:

FakeHeteroDataset (#graphs=20):
+------------+----------+----------+
|            |   #nodes |   #edges |
|------------+----------+----------|
| mean       |   3945.4 |  49149.2 |
| std        |    247.8 |   3481.2 |
| min        |   3300   |  40440   |
| quantile25 |   3849   |  47016.8 |
| median     |   3974   |  49820   |
| quantile75 |   4129   |  51367.8 |
| max        |   4207   |  53761   |
+------------+----------+----------+
Number of nodes per node type:
+------------+--------+--------+--------+--------+
|            |    #v0 |    #v1 |    #v2 |    #v3 |
|------------+--------+--------+--------+--------|
| mean       |  980.5 |  994.8 |  944.7 | 1025.4 |
| std        |  140.7 |  123.8 |  123.9 |  123.7 |
| min        |  765   |  770   |  754   |  753   |
| quantile25 |  862.2 |  895   |  863   |  936.2 |
| median     |  979   | 1005   |  915   | 1067   |
| quantile75 | 1068.8 | 1064.8 | 1036   | 1103.2 |
| max        | 1235   | 1197   | 1201   | 1225   |
+------------+--------+--------+--------+--------+
Number of edges per edge type:
+------------+-----------------------+-----------------------+-----------------------+-----------------------+-----------------------+
|            |   #('v0', 'e0', 'v2') |   #('v2', 'e0', 'v0') |   #('v1', 'e0', 'v0') |   #('v3', 'e0', 'v0') |   #('v1', 'e0', 'v3') |
|------------+-----------------------+-----------------------+-----------------------+-----------------------+-----------------------|
| mean       |                9749.5 |                9400.5 |                9897   |               10203   |                9899.2 |
| std        |                1395.8 |                1229.9 |                1230.2 |                1230.6 |                1228.8 |
| min        |                7613   |                7502   |                7650   |                7498   |                7657   |
| quantile25 |                8577.2 |                8589.5 |                8899.8 |                9315.2 |                8901.5 |
| median     |                9732   |                9108   |               10014   |               10611   |                9978   |
| quantile75 |               10624.5 |               10307.8 |               10584.8 |               10971.5 |               10597   |
| max        |               12278   |               11935   |               11909   |               12186   |               11908   |
+------------+-----------------------+-----------------------+-----------------------+-----------------------+-----------------------+

For datasets that haven't node_types or edge_types defined or calling from_dataset method with parameter per_type_breakdown=False output will be the same as before the change.

[codecov]

Codecov Report

Merging #7203 (c4b2d49) into master (b07a84f) will decrease coverage by 0.43%.
The diff coverage is 100.00%.

❗ Current head c4b2d49 differs from pull request most recent head deb655c. Consider uploading reports for the commit deb655c to get more accurate results

@@            Coverage Diff             @@
##           master    #7203      +/-   ##
==========================================
- Coverage   92.03%   91.61%   -0.43%     
==========================================
  Files         437      437              
  Lines       24111    24140      +29     
==========================================
- Hits        22191    22116      -75     
- Misses       1920     2024     +104     

Impacted Files	Coverage Δ
torch_geometric/data/summary.py	98.75% <100.00%> (+0.83%)	⬆️

... and 20 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more