Under review at INFORMS Journal on Computing.
* Corresponding author.
- arXiv: 2608.19231
- Repository: GitHub
TorchDCM is a unified PyTorch-native package for discrete choice modeling. It compiles choice data and model specifications into a likelihood engine for estimation, inference, prediction, and structured reporting on CPU and CUDA devices. The package covers multinomial, nested, mixed, ordered, latent-variable, and panel likelihoods, together with constrained parameters, covariance estimation, willingness-to-pay analysis, elasticities, and extensible likelihood components. Aligned synthetic and real-data experiments compare TorchDCM with seven other estimation packages. TorchDCM completes all 45 synthetic cases, runs fastest in every comparable synthetic case, and meets the prespecified final-log-likelihood tolerance whenever at least two comparable solutions are available.
