pycmtensor 1.6.1

Python Tensor based package for discrete choice modelling.

Tensor-based choice modelling estimation Python package

Welcome

PyCMTensor is a tensor-optimized discrete choice model estimation Python library package, written with optimization compilers to speed up estimation of large datasets, simulating very large mixed logit models or implementing neural network functions into utility equations in choice models.

Currently, PyCMTensor can be used to fully specify Multinomial Logit and Mixed Logit models, estimate and generate statistical tests, using optimized tensor operations via Aesara tensor libraries.

Key features

Main features:

• Interpretable and customizable utility specification syntax
• Perform statistical tests and generate var-covar matrices for taste parameters.
• Fast execution of model estimation including of simulation based methods, e.g. Mixed Logit models
• Model estimation with 1st order (Stochastic Gradient Descent) or 2nd order methods (BFGS)
• Specifying neural nets with weight and bias parameters inside a utility function [TODO]

While other choice modelling estimation software in Python are available, e.g. Biogeme, xlogit, PyLogit, etc., PyCMTensor strives to fully implement deep learning based methods written in the same syntax format as Biogeme. Different software programs may occasionally vary in their behaviour and estimation results. The following are some of the key differences between PyCMTensor and other choice modelling estimation packages:

Documentation

See documentation at https://mwong009.github.io/pycmtensor/