What's your MAE? Share your model diagnostics

[m-zenker]

The rolling MAE sensors (sensor.energy_forecast_mae_7d, mae_30d, relative_mae_7d) are the best way to gauge whether your model is well-fitted for your household. I'm curious how others compare — hardware, setup complexity, and time running all affect accuracy.

My numbers (to give you a reference):

• 7d MAE: 0.16 kWh/h
• 30d MAE: 0.20 kWh/h
• Relative 7d MAE: 60% — grid import averages ~0.26 kWh/h and even a small absolute error looks large as a percentage. Absolute MAE is the more honest metric.
• Running since: October 2025 (~8 months of training data)
• Hardware: Raspberry Pi 5 8 GB (LightGBM, not sklearn fallback)
• Optional features enabled: heat pump sub-sensor, thermal pressure, DHW pressure, daily regime clustering, EV detection, solar + battery target correction
  Share whatever you have — even a rough number is useful. If your MAE feels high, mention your setup and we can troubleshoot.