How to Package Your Model for Competition Success: The Three‑Step ‘Ancient Kung Fu’ Method

In a follow‑up to a previous story‑telling post, the author shares a concrete methodology for turning an ordinary competition model into a compelling narrative that judges will appreciate. The approach is framed as three “ancient kung‑fu” techniques.

First Axe – "Pretty Clothes"

The goal is to create attractive visualizations that showcase the overall pipeline or model architecture. For deep‑learning models, high‑quality diagrams from papers are recommended. When using tree‑based models, a clean pipeline diagram can replace missing architecture visuals. A useful collection of visual assets is available at https://github.com/dair-ai/ml-visuals. The author also cites a KDD competition team (88VIP) whose poster‑style diagram effectively communicates the workflow.

Second Axe – "Good Song"

This step focuses on highlighting a novel optimization or insight that makes the solution stand out. Two strategies are described:

Bottom‑up : Identify a problem (e.g., cold‑start in recommendation) and develop a solution, then emphasize this discovery during the presentation.

Top‑down : Start with the best performing model, trace back to the underlying problem it solves, and then showcase how your enhancements improve upon it.

Both approaches should be supported by clear ablation or performance graphs that draw the judges' attention to the “eye‑catching” improvement.

Third Axe – "Rich Breakfast"

Here the emphasis is on thorough, credible experimental comparisons. Provide a suite of baseline and variant experiments that demonstrate the robustness of your solution. The author references an external article titled “Doing Competitions Like Governing a Country” for additional examples. Even if the model is not groundbreaking, a well‑structured set of experiments conveys diligence and scientific rigor.

Finally, the three axes are summarized:

Pretty Clothes – clear, attractive visualizations of the model or pipeline.

Good Song – a highlighted, novel optimization presented with a compelling narrative.

Rich Breakfast – extensive, trustworthy experimental results that justify the chosen solution.

Applying these steps helps transform a plain model into a story that judges can easily follow and appreciate.