azencot-group/KoVAE — reverse-engineered prompt

Build me a Python research project for KoVAE, a Koopman VAE for generating and modeling regular and irregular time series data. I want to be able to set up the environment, train on the included stock and energy data, and also generate a simple sine dataset for testing.

Please include easy command line scripts so I can run regular training by choosing the dataset and weights for KL and prediction loss. Also add an irregular training option where I can choose the missing data rate, like 0.3, 0.5, or 0.7, and set the input channel count. If irregular data needs preprocessing, make it save the processed version so it doesn’t repeat the slow work every time.

Keep it close to the KoVAE paper idea, with a clear model folder, dataset loading helpers, metrics, and simple example shell commands for sine and stock experiments. Add a README that tells me exactly how to create the conda environment and run the training.