OpenSynth New England — CVAE Weights
Conditional Variational Autoencoder for generating synthetic AMI load profiles conditioned on New England housing archetypes, heating fuel, weather, and DER.
Usage
pip install opensynth-newengland[cvae]
# Download weights to src/opensynth_ne/weights/
from opensynth_ne import generate_day
df, summary = generate_day(archetype="colonial", fuel="oil", month=0)
Architecture
- 948k parameters
- 6-channel output: base, behav, heat, cool, ev, solar (96 timesteps each)
- 32-dim latent space, 25-dim conditioning vector
- Trained with cyclical beta annealing + free bits (0.5 nats/dim)
Training
- 500k synthetic profiles from v0 mock model
- 200 epochs, early stopped at 70
- Val reconstruction MSE: 0.2225
Repository
- Downloads last month
- 2
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support