Workshop Started
New cohort working with gradient boosting
Our latest group just started exploring how XGBoost and LightGBM handle market regime detection. They're comparing performance across different asset classes and timeframes.
View program
We're tracking everything that matters in the space where machine learning meets trading decisions. Real applications, new approaches, and insights you can actually use in your work.
Our latest group just started exploring how XGBoost and LightGBM handle market regime detection. They're comparing performance across different asset classes and timeframes.
View program
We're opening spots for people interested in building meta-learners that combine different prediction approaches. Focus is on practical implementation with real trading constraints.
Get detailsParticipants discovered that feature importance rankings shift dramatically across different market conditions. They built separate models for high and low volatility regimes, improving prediction accuracy by focusing on regime-specific indicators.
A recent group tested how bootstrap aggregating handles fundamental data versus technical indicators. Turns out combining both types through bagged decision trees reduced overfitting issues that plagued their single-model approaches.
Workshop participants compared static versus dynamic ensemble weights. Dynamic reweighting based on recent performance helped adapt to changing market conditions, though it added complexity to the implementation process.
Market events create naturally imbalanced data. Workshop groups experimented with SMOTE variations and cost-sensitive learning in ensemble contexts. Results showed that proper handling of class imbalance matters more than the specific ensemble technique used.
Regular k-fold validation breaks when you're dealing with sequential financial data. Participants learned walk-forward analysis and expanding window approaches that respect temporal ordering while still enabling proper model evaluation.
Get access to working implementations of voting classifiers, stacking regressors, and boosting algorithms applied to financial data. Includes preprocessing pipelines and performance evaluation frameworks.
Explore programWant to know more about how we structure workshops or what participants actually work on? We're happy to explain our methods and discuss whether this learning style fits what you're looking for.
Get in touch