EnergyTrader | ML-Driven Market Platform

Model Training

Train the HistGradientBoosting price forecasting model for DAM / RTM / TAM

No data found for DAM. Run a scrape first.

Split & Shuffle

Training window (most recent N days)

All data

Test Size (held-out validation)

20%

Shuffle training data

Hyperparameters (HistGradientBoosting)

Max Iterations?Total number of boosting rounds. Early stopping will halt before this if validation loss stops improving.Max Depth?Maximum depth of each decision tree. Deeper trees capture more complex patterns but are prone to overfitting. Typical sweet spot: 4–12.Learning Rate?Shrinkage applied to each tree's contribution. Lower values generalise better but require more iterations.Min Samples Leaf?Minimum samples in a leaf node. Higher values smooth the model and reduce overfitting.L2 Regularisation?L2 penalty on leaf values. Increase if the model overfits historical price spikes.Max Bins?Bins for discretising continuous features. 255 is the maximum.N Iter No Change?Iterations with no improvement before early stopping triggers.Val Fraction?Fraction of training data held out for early-stopping validation.

Early stopping

Data Enrichment

Checking enrichment status…

Fetches weather (wind, solar, cloud) from Open-Meteo ERA5 archive and computes holiday flags. Required before training with weather / holiday features.

Feature Engineering

Extra Lag Days?Additional historical days as features. E.g. "2,3,14" adds price from 2, 3, and 14 days ago. Day-1 lag is always included.Rolling Windows (days)?Window sizes for rolling mean and std features. E.g. "7,14" adds 7-day and 14-day rolling averages.EMA Span?Window size for the exponential moving average feature.

D/S ratioPrice momentumEMA

New features (run enrichment first)

Weatherwind, solar, cloud

Holidaysis_holiday, days_to_holiday

Month cyclicsin/cos encoding

Price range7-day high−low spread

Summer signalis_summer + max_temp_7d

Auto-Tuning (Hyperparameter Search)

Enable auto-tuning

MethodIterations?Random parameter combinations to evaluate. Ignored for Grid Search.CV Folds?Cross-validation folds. More folds = more robust estimate but slower.Scoring