Autoresearch Dashboard

Karpathy-Style Autonomous ML Research · Electricity Price Forecasting · DE_LU
0.46
Best MAE (€/MWh)
54
Experiments
17
Kept
20030
BESS Profit (€)
-98.7%
MAE Reduction

MAE Progression — All 54 Experiments

Architecture Evolution

Experiment Outcomes

Key Research Insights

Prophet → GBM

Switching from Prophet (MAE 36.63) to gradient boosting reduced error by 96%. Tabular models dominate for this task.

Ensemble Power

Single model (1.12) → 2-model (1.02) → 3-model (0.70) → 7-model (0.46). Each diversity source helps.

Seed Diversity

Adding HGB models with different seeds (42/123/777) provided cheap diversity. More seeds hit diminishing returns at 3.

Quantile Trick

HGB with q0.45 and q0.55 loss added asymmetric diversity that simple seed changes couldn't achieve. Final breakthrough.

Weight Optimization

Scipy Nelder-Mead on val MAE found optimal weights. Equal weights ≈ optimized for similar-strength models, but matters for diverse ones.

Dead Ends

Neural nets (MLP), stacking, Fourier features, feature interactions, target encoding — all failed. GBMs handle these internally.

Experiment Log