Estimation Tools Review

The purpose of this page is to review available Python estimation tools in order to select one for prototyping estimation integration. The activitysim sub-models to be integrated and estimated at this point in time are auto ownership, work location, and tour mode choice. Auto ownership is a multinomial logit model (MNL) with few alternatives, work location is an MNL model with size terms and many alternatives, and tour mode choice is a nested logit (NL) model with a few alternatives and a significant amount of data.

Feature	larch	PandasBiogeme	pylogit	choicemodels
Includes MNL	x	x	x	x
Includes NL	x	x	x
Efficient handling of large alternative sets	x	x		x
Active development	x			x
Sufficient user community	x	x	x	x
Sufficient documentation	x	x	x	x
Travel modeling industry maintainer	x	x
Easy-to-use, focus on practical models	x		x	x

We've decided to use larch for prototyping because it supports MNL and NL, large alternative sets, is designed for linear models and is therefore straightforward to use, is active, is maintained by a member of the industry, has good documentation, and has a sufficient user base.