2026-01-31
By using the GNN, we can evaluate any hypothetical lineup. Including combinations across teams, where little to no historical line data may exist.
In my previous post, we looked at all possible combinations of lines and pairings on a team's roster. But, where this is most interesting, is looking at combinations across teams. Typically this would be hamstrung by a lack of data; there's limited data on how players from different teams would play together. With the Olympics approaching, this presents a fascinating challenge. Searching the full pool of eligible players produces hundreds of thousands of potential combinations.
Below, I have used the GNN embeddings to score every possible line and pairing for the Canadian, Finnish, Swedish, and American roster using skater data from each of 2024 and 2025. For the Canadian and American teams, exclusively, I also limited the candidate lines and pairings to players that were invited to their Olympic camps.
While the GNN provides interesting output, it is currently limited by the data. I have only utilized data from individual seasons; which is how the data is aggregated and available from MoneyPuck1.
In general, international rosters are deferential to a players body of work. Because the model utilizes single season data, it is responsive to players having a great year. We can see this impact on the Canadian teams utilizing all available player data from 2025: it's dominated by Colorado defensemen and Utah forwards.
An ideal evolution of this approach would utilze a rolling window of data spanning multiple seasons of games played, potentially weighing more recent games higher. This would better capture both sustained elite performance and current form, provide a more balanced view that aligned with how national team rosters are built.
In the next post, we'll use our GNN embeddings to produce a ranking of the most impactful players in the league.
All posts in this series:
1. The Line Blender: Optimizing Lineups Using MoneyPuck's Expected Goals Percentage (xG%)
2. The Line Blender: Embedding Line Performance Using a GNN
3. The Line Blender: Using a GNN To Produce Olympic Rosters
4. The Line Blender: Using GNN Embeddings for Player Rankings
5. The Line Blender: Olympic Lineups with Announced Rosters
6. The Line Blender: Hypothetical Russian Olympic Lineups