https://galliard7.github.io/tags/kaggle/Recent content in Kaggle on Galliard7Hugoen-usMon, 15 Jan 2024 00:00:00 +0000https://galliard7.github.io/projects/pii-data-detection/Mon, 15 Jan 2024 00:00:00 +0000https://galliard7.github.io/projects/pii-data-detection/<h2 id="overview">Overview</h2> <p>Detecting personally identifiable information (PII) in student essays using named entity recognition. The challenge required identifying and classifying PII tokens across thousands of documents with high precision and recall.</p> <h2 id="approach">Approach</h2> <ul> <li><strong>Synthetic data generation</strong> to augment limited training data with realistic PII patterns</li> <li><strong>DeBERTa-based NER models</strong> fine-tuned for token-level PII classification</li> <li><strong>ONNX optimization</strong> for inference speed without sacrificing accuracy</li> <li><strong>Ensemble strategy</strong> combining multiple model checkpoints</li> </ul> <h2 id="result">Result</h2> <p>121/2048 🥉</p> <ul> <li><strong>Public leaderboard:</strong> 0.970</li> <li><strong>Private leaderboard:</strong> 0.956</li> </ul> <h2 id="links">Links</h2> <ul> <li><a href="https://github.com/Galliard7/pii-data-detection">GitHub Repository</a></li> </ul>https://galliard7.github.io/projects/llm-science-exam/Sun, 15 Oct 2023 00:00:00 +0000https://galliard7.github.io/projects/llm-science-exam/<h2 id="overview">Overview</h2> <p>Answering difficult science questions using large language models. The competition required selecting correct answers from multiple-choice science questions spanning physics, chemistry, biology, and other domains.</p> <h2 id="result">Result</h2> <p>194/2664 🥉</p> <h2 id="links">Links</h2> <ul> <li><a href="https://github.com/Galliard7/llm-science-exam">GitHub Repository</a></li> </ul>https://galliard7.github.io/projects/amex-default-prediction/Thu, 15 Jun 2023 00:00:00 +0000https://galliard7.github.io/projects/amex-default-prediction/<h2 id="overview">Overview</h2> <p>Predicting credit card default probability for American Express customers using anonymized transaction and account features. A large-scale tabular competition with heavy feature engineering requirements.</p> <h2 id="approach">Approach</h2> <ul> <li>Extensive feature engineering over time-series transaction histories</li> <li>Gradient boosting models (LightGBM, XGBoost, CatBoost)</li> <li>Aggregation features: rolling statistics, lag features, trend indicators</li> <li>Careful handling of missing values and categorical encodings</li> </ul> <h2 id="result">Result</h2> <p>287/4874 🥉</p> <h2 id="links">Links</h2> <ul> <li><a href="https://github.com/Galliard7/amex-default-prediction">GitHub Repository</a></li> </ul>https://galliard7.github.io/projects/jane-street-market-prediction/Wed, 15 Mar 2023 00:00:00 +0000https://galliard7.github.io/projects/jane-street-market-prediction/<h2 id="overview">Overview</h2> <p>Predicting profitable trading opportunities from anonymized financial market data. The competition required building models that could identify actionable signals in noisy, high-dimensional market features.</p> <h2 id="approach">Approach</h2> <ul> <li>Feature selection and denoising on anonymized market signals</li> <li>Gradient boosting and neural network ensembles</li> <li>Custom utility-based optimization aligned with competition metric</li> <li>Time-aware validation to avoid lookahead bias</li> </ul> <h2 id="links">Links</h2> <ul> <li><a href="https://github.com/Galliard7/jane-street-market-prediction">GitHub Repository</a></li> </ul>https://galliard7.github.io/projects/ventilator-pressure-prediction/Mon, 15 Nov 2021 00:00:00 +0000https://galliard7.github.io/projects/ventilator-pressure-prediction/<h2 id="overview">Overview</h2> <p>Predicting airway pressure in mechanically ventilated patients. The challenge simulated a ventilator connected to a sedated patient&rsquo;s lung, requiring models to predict pressure time series given control inputs.</p> <h2 id="result">Result</h2> <p>185/2605 🥉</p> <h2 id="links">Links</h2> <ul> <li><a href="https://github.com/Galliard7/ventilator-pressure-prediction">GitHub Repository</a></li> </ul>https://galliard7.github.io/projects/google-research-football/Tue, 15 Sep 2020 00:00:00 +0000https://galliard7.github.io/projects/google-research-football/<h2 id="overview">Overview</h2> <p>Building AI agents that play 11v11 simulated football. Agents receive game observations (player positions, ball state, game mode) and return actions, competing head-to-head on Kaggle&rsquo;s evaluation servers with Elo-style rating.</p> <h2 id="approach">Approach</h2> <ul> <li><strong>Rule-based tactical foundation</strong> — &ldquo;marauding wingers&rdquo; formation: wide players sprint down flanks and deliver crosses into the box</li> <li><strong>Zone-based decision architecture</strong> — field divided into zones (defensive third, wing corridors, crossing range, shooting range) with different behaviors per zone</li> <li><strong>Opponent-aware mechanics</strong> — proximity detection for context-sensitive decisions: sprint in open space, dribble under pressure, pass when crowded</li> <li><strong>Goalkeeper exploitation</strong> — specific logic to detect when the opposing keeper is out of position and trigger long-range shots</li> <li><strong>Sprint/dribble state machine</strong> — manages action mode based on field position and opponent proximity</li> </ul> <h2 id="result">Result</h2> <p>61/1138 🥉</p>