Chinese version Solution:
Competition Link:
Code Repository:
First and foremost, we would like to express our heartfelt gratitude to CrunchDAO and ADIA Lab for organizing this competition. Our sincere thanks go to all the staff involved for their hard work in creating such an exciting and challenging event. Additionally, I would like to thank my university, ShanghaiTech, for promoting this competition and my two classmates for informing me about it. Their kindness and support enabled me to participate in this competition.
This solution was a collaborative effort by Mutian Hong from ShanghaiTech University and Guoqin Gu from Xiamen University.
We achieved balanced accuracy of 73.43% on the public leaderboard and 74.06% in the out-of-sample stage.
Contract Us:
In this competition, our objective was to identify causal relationships. Specifically, given a causal graph with a known relationship X→Y, where X and Y are two nodes, the graph contains additional nodes named as numbers from 0 to 9. These nodes might play different roles in the causal relationship between X and Y. For instance, in the diagram above, node 1 is a common cause of X and Y, node 0 is their common effect, while node 2 is independent of both X and Y.
The organizers provided observational data generated from causal graphs following specific rules. The left side of the diagram above shows an example of 1,000 observations. The task was framed as a classification problem to determine the type of relationship between other nodes (v) and X, Y—for example, whether v→X. If this edge exists, the label is 1; otherwise, it is 0.
Upon further analysis, there are three possible relationships between v and X: v→X, X→v, or independence. Similarly, there are three possibilities for v and Y. Combining these, we arrive at 9 possible relationships. However, one configuration (Y→v→X) is impossible due to the acyclic nature of causal graphs. Hence, we defined this as an 8-class classification problem with the following categories:
| Role | Relationship |
|---|---|
| Confounder | X←v→Y |
| Collider | X→v←Y |
| Mediator | X→v→Y |
| Cause of X | v→X |
| Cause of Y | v→Y |
| Effect of X | X→v |
| Effect of Y | Y→v |
| Independent | v |