Skip to content

add benchmark results and description update to docs #41

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pjhartout merged 3 commits into from
Oct 7, 2025
Merged

add benchmark results and description update to docs #41

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
018428a
add benchmark results and description update to docs
pjhartout Oct 7, 2025
259626e
transpose table, remove VUN
pjhartout Oct 7, 2025
d807770
Update version to 1.0.1, remove redundant section from README, and en…
pjhartout Oct 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
66 changes: 66 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,3 +172,69 @@ for metric in tqdm(metrics):
generated,
)
```
## Example Benchmark

The following results mirror the tables from our paper. Bold indicates best, and underlined indicates second-best. Values are multiplied by 100 for legibility. Standard deviations are obtained with subsampling using `StandardPGDInterval` and `MoleculePGDInterval`. Specific parameters are discussed in the paper.

<table>
<thead>
<tr>
<th>Method</th>
<th style="text-align:right;">Planar-L</th>
<th style="text-align:right;">Lobster-L</th>
<th style="text-align:right;">SBM-L</th>
<th style="text-align:right;">Proteins</th>
<th style="text-align:right;">Guacamol</th>
<th style="text-align:right;">Moses</th>
</tr>
</thead>
<tbody>
<tr>
<td>AutoGraph</td>
<td style="text-align:right;"><strong>34.0 ± 1.8</strong></td>
<td style="text-align:right;"><u>18.0 ± 1.6</u></td>
<td style="text-align:right;"><strong>5.6 ± 1.5</strong></td>
<td style="text-align:right;"><strong>67.7 ± 7.4</strong></td>
<td style="text-align:right;"><u>22.9 ± 0.5</u></td>
<td style="text-align:right;"><strong>29.6 ± 0.4</strong></td>
</tr>
<tr>
<td>AutoGraph*</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;"><strong>10.4 ± 1.2</strong></td>
<td style="text-align:right;">—</td>
</tr>
<tr>
<td>DiGress</td>
<td style="text-align:right;">45.2 ± 1.8</td>
<td style="text-align:right;"><strong>3.2 ± 2.6</strong></td>
<td style="text-align:right;"><u>17.4 ± 2.3</u></td>
<td style="text-align:right;">88.1 ± 3.1</td>
<td style="text-align:right;">32.7 ± 0.5</td>
<td style="text-align:right;"><u>33.4 ± 0.5</u></td>
</tr>
<tr>
<td>GRAN</td>
<td style="text-align:right;">99.7 ± 0.2</td>
<td style="text-align:right;">85.4 ± 0.5</td>
<td style="text-align:right;">69.1 ± 1.4</td>
<td style="text-align:right;">89.7 ± 2.7</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
</tr>
<tr>
<td>ESGG</td>
<td style="text-align:right;"><u>45.0 ± 1.4</u></td>
<td style="text-align:right;">69.9 ± 0.6</td>
<td style="text-align:right;">99.4 ± 0.2</td>
<td style="text-align:right;"><u>79.2 ± 4.3</u></td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
</tr>
</tbody>
</table>

<sub>* AutoGraph* denotes a variant that leverages additional training heuristics as described in the paper.</sub>
258 changes: 62 additions & 196 deletions docs/index.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,201 +92,67 @@ PGD and its motivation are described in more detail in the paper and API docs.

### Benchmarking snapshot

The table below shows an example benchmark generated with this library across multiple datasets and models. Values illustrate typical outputs from the implemented metrics (VUN, PGD, and PGD subscores). For completeness, this library and our paper also implements and provides various MMD estimates on the datasets below.

<style>
table {
font-size: 85%;
border-collapse: collapse;
}
th, td {
text-align: center;
padding: 4px 6px;
border: 1px solid #ddd;
}
th:first-child, td:first-child {
text-align: left;
}
th:nth-child(2), td:nth-child(2) {
text-align: left;
}
</style>
The table below shows an example benchmark generated with this library across multiple datasets and models. Values illustrate typical outputs from the newly proposed PolyGraph Discrepancy. For completeness, this library and our paper also implements and provides various MMD estimates on the datasets below. Values are scaled by 100 for legibility and subsampling is used to obtain standard deviations (using `StandardPGDInterval` and `MoleculePGDInterval`). More details are provided in our paper.

<table>
<thead>
<tr>
<th rowspan="2">Dataset</th>
<th rowspan="2">Model</th>
<th rowspan="2">VUN (↑)</th>
<th rowspan="2">PGD (↓)</th>
<th colspan="6">PGD subscores</th>
</tr>
<tr>
<th>Clust. (↓)</th>
<th>Deg. (↓)</th>
<th>GIN (↓)</th>
<th>Orb5. (↓)</th>
<th>Orb4. (↓)</th>
<th>Eig. (↓)</th>
</tr>
</thead>
<tbody>

<!-- Planar-L -->
<tr><td rowspan="4"><b>Planar-L</b></td>
<td>AutoGraph</td>
<td><i>85.1</i></td>
<td><b>34.0 ± 1.8</b></td>
<td><b>7.0 ± 2.9</b></td>
<td><b>7.8 ± 3.2</b></td>
<td><b>8.8 ± 3.0</b></td>
<td><b>34.0 ± 1.8</b></td>
<td><b>28.5 ± 1.5</b></td>
<td><b>26.9 ± 2.3</b></td></tr>
<tr><td>DiGress</td>
<td>80.1</td>
<td>45.2 ± 1.8</td>
<td>24.8 ± 2.0</td>
<td>23.3 ± 1.2</td>
<td><i>29.0 ± 1.1</i></td>
<td>45.2 ± 1.8</td>
<td><i>40.3 ± 1.8</i></td>
<td>39.4 ± 2.0</td></tr>
<tr><td>GRAN</td>
<td>1.6</td>
<td>99.7 ± 0.2</td>
<td>99.3 ± 0.2</td>
<td>98.3 ± 0.3</td>
<td>98.3 ± 0.3</td>
<td>99.7 ± 0.1</td>
<td>99.2 ± 0.2</td>
<td>98.5 ± 0.4</td></tr>
<tr><td>ESGG</td>
<td><b>93.9</b></td>
<td><i>45.0 ± 1.4</i></td>
<td><i>10.9 ± 3.2</i></td>
<td><i>21.7 ± 3.0</i></td>
<td>32.9 ± 2.2</td>
<td><i>45.0 ± 1.4</i></td>
<td>42.8 ± 1.9</td>
<td><i>29.6 ± 1.6</i></td></tr>

<!-- Lobster-L -->
<tr><td rowspan="4"><b>Lobster-L</b></td>
<td>AutoGraph</td>
<td><i>83.1</i></td>
<td><i>18.0 ± 1.6</i></td>
<td>4.2 ± 1.9</td>
<td><i>12.1 ± 1.6</i></td>
<td><i>14.8 ± 1.5</td>
<td><i>18.0 ± 1.6</i></td>
<td><i>16.1 ± 1.6</i></td>
<td><i>13.0 ± 1.1</i></td></tr>
<tr><td>DiGress</td>
<td><b>91.4</b></td>
<td><b>3.2 ± 2.6</b></td>
<td><i>2.0 ± 1.3</i></td>
<td><b>1.2 ± 1.5</b></td>
<td><b>2.3 ± 2.0</b></td>
<td><b>3.0 ± 3.1</b></td>
<td><b>4.5 ± 2.3</b></td>
<td><b>1.3 ± 1.1</b></td></tr>
<tr><td>GRAN</td>
<td>41.3</td>
<td>85.4 ± 0.5</td>
<td>20.8 ± 1.1</td>
<td>77.1 ± 1.2</td>
<td>79.8 ± 0.6</td>
<td>85.4 ± 0.5</td>
<td>85.0 ± 0.6</td>
<td>69.8 ± 1.2</td></tr>
<tr><td>ESGG</td>
<td>70.9</td>
<td>69.9 ± 0.6</td>
<td><b>0.0 ± 0.0</b></td>
<td>63.4 ± 1.1</td>
<td>66.8 ± 1.0</td>
<td>69.9 ± 0.6</td>
<td>66.0 ± 0.6</td>
<td>51.7 ± 1.8</td></tr>

<!-- SBM-L -->
<tr><td rowspan="4"><b>SBM-L</b></td>
<td>AutoGraph</td>
<td><b>85.6</b></td>
<td><b>5.6 ± 1.5</b></td>
<td><b>0.3 ± 0.6</b></td>
<td><b>6.2 ± 1.4</b></td>
<td><b>6.3 ± 1.3</td>
<td><b>3.2 ± 2.2</td>
<td><b>4.4 ± 2.0</td>
<td><b>2.5 ± 2.2</td></tr>
<tr><td>DiGress</td>
<td><i>73.0</i></td>
<td><i>17.4 ± 2.3</td>
<td><i>5.7 ± 2.8</td>
<td><i>8.2 ± 3.3</td>
<td><i>13.8 ± 1.7</td>
<td><i>17.4 ± 2.3</td>
<td><i>14.8 ± 2.5</td>
<td><i>8.7 ± 3.0</td></tr>
<tr><td>GRAN</td>
<td>21.4</td>
<td>69.1 ± 1.4</td>
<td>50.2 ± 1.9</td>
<td>58.6 ± 1.4</td>
<td>69.1 ± 1.4</td>
<td>65.7 ± 1.3</td>
<td>62.8 ± 1.3</td>
<td>55.9 ± 1.5</td></tr>
<tr><td>ESGG</td>
<td>10.4</td>
<td>99.4 ± 0.2</td>
<td>97.9 ± 0.5</td>
<td>97.5 ± 0.6</td>
<td>98.3 ± 0.4</td>
<td>96.8 ± 0.4</td>
<td>89.2 ± 0.7</td>
<td>99.4 ± 0.2</td></tr>

<!-- Proteins -->
<tr><td rowspan="4"><b>Proteins</b></td>
<td>AutoGraph</td>
<td>–</td>
<td><b>67.7 ± 7.4</b></td>
<td><i>47.7 ± 5.7</i></td>
<td><i>31.5 ± 8.5</td>
<td><i>45.3 ± 5.1</td>
<td><b>67.7 ± 7.4</td>
<td><b>47.4 ± 7.0</td>
<td>53.2 ± 6.9</td></tr>
<tr><td>DiGress</td>
<td>–</td>
<td>88.1 ± 3.1</td>
<td><b>36.1 ± 4.3</b></td>
<td><b>29.2 ± 5.0</b></td>
<td><b>23.2 ± 5.3</b></td>
<td>88.1 ± 3.1</td>
<td><i>60.8 ± 3.6</i></td>
<td><b>23.4 ± 11.8</b></td></tr>
<tr><td>GRAN</td>
<td>–</td>
<td>89.7 ± 2.7</td>
<td>86.0 ± 2.0</td>
<td>70.6 ± 3.1</td>
<td>71.5 ± 3.0</td>
<td>90.4 ± 2.4</td>
<td>84.4 ± 3.3</td>
<td>76.7 ± 4.7</td></tr>
<tr><td>ESGG</td>
<td>–</td>
<td><i>79.2 ± 4.3</td>
<td>58.2 ± 3.6</td>
<td>54.0 ± 3.6</td>
<td>57.4 ± 4.1</td>
<td><i>80.2 ± 3.1</td>
<td>72.5 ± 3.0</td>
<td><i>24.3 ± 11.0</td></tr>

</tbody>
</table>
<thead>
<tr>
<th>Method</th>
<th style="text-align:right;">Planar-L</th>
<th style="text-align:right;">Lobster-L</th>
<th style="text-align:right;">SBM-L</th>
<th style="text-align:right;">Proteins</th>
<th style="text-align:right;">Guacamol</th>
<th style="text-align:right;">Moses</th>
</tr>
</thead>
<tbody>
<tr>
<td>AutoGraph</td>
<td style="text-align:right;"><strong>34.0 ± 1.8</strong></td>
<td style="text-align:right;"><u>18.0 ± 1.6</u></td>
<td style="text-align:right;"><strong>5.6 ± 1.5</strong></td>
<td style="text-align:right;"><strong>67.7 ± 7.4</strong></td>
<td style="text-align:right;"><u>22.9 ± 0.5</u></td>
<td style="text-align:right;"><strong>29.6 ± 0.4</strong></td>
</tr>
<tr>
<td>AutoGraph*</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;"><strong>10.4 ± 1.2</strong></td>
<td style="text-align:right;">—</td>
</tr>
<tr>
<td>DiGress</td>
<td style="text-align:right;">45.2 ± 1.8</td>
<td style="text-align:right;"><strong>3.2 ± 2.6</strong></td>
<td style="text-align:right;"><u>17.4 ± 2.3</u></td>
<td style="text-align:right;">88.1 ± 3.1</td>
<td style="text-align:right;">32.7 ± 0.5</td>
<td style="text-align:right;"><u>33.4 ± 0.5</u></td>
</tr>
<tr>
<td>GRAN</td>
<td style="text-align:right;">99.7 ± 0.2</td>
<td style="text-align:right;">85.4 ± 0.5</td>
<td style="text-align:right;">69.1 ± 1.4</td>
<td style="text-align:right;">89.7 ± 2.7</td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
</tr>
<tr>
<td>ESGG</td>
<td style="text-align:right;"><u>45.0 ± 1.4</u></td>
<td style="text-align:right;">69.9 ± 0.6</td>
<td style="text-align:right;">99.4 ± 0.2</td>
<td style="text-align:right;"><u>79.2 ± 4.3</u></td>
<td style="text-align:right;">—</td>
<td style="text-align:right;">—</td>
</tr>
</tbody>
</table>

<sub>* AutoGraph* denotes a variant that leverages additional training heuristics as described in the paper.</sub>
2 changes: 1 addition & 1 deletion pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"

[project]
name = "polygraph-benchmark"
version = "1.0.0"
version = "1.0.1"
description = "Evaluation benchmarks for graph generative models"
readme = "README.md"
authors = [
Expand Down