TorchDCM is a unified PyTorch-native package for estimating and analyzing discrete choice models, co-developed with Baichuan.
- Repository: github.com/mbc96325/torchdcm
The package provides vectorized likelihoods built around PyTorch tensors and automatic differentiation, with support for multinomial, nested, cross-nested, mixed, latent-class, ordered, hybrid, and related choice models. It also includes classic, robust, and cluster covariance estimates, as well as WTP and elasticity helpers.