partworth: hierarchical Bayes for choice-based conjoint in PyMC

[Koji_M]

Hi all. I am Koji Miyazawa, a data scientist based in Japan. I work with healthcare data and Bayesian models.

I saw a few questions here about hierarchical models for choice-based conjoint (CBC) in PyMC, for example:

Choice-Based Conjoint Data Using A Hierarchical Logit Model

Hello, I am trying to write an implementation for this model for the camera dataset bayesm Overview, my model already runs for 2 clients and for 332 clients it takes to long to start the sampling. Can you give me a hint, so that I can write the model in a more efficient way? This is my code: import json import numpy as np import pymc as pm import arviz as az import matplotlib.pyplot as plt import seaborn as sns from tqdm import tqdm # Import tqdm import networkx as nx figsize = (16, 10) plt.…

I wrote a small open-source package called partworth.
GitHub - kojikojikojikoji/partworth: Hierarchical Bayes choice-based conjoint analysis in Python (PyMC), with Sawtooth-style outputs and a market simulator · GitHub (pip install partworth)

It has:

• checks for long-format choice data, tasks of different sizes and a “none” option
• hierarchical MNL with correlated coefficients (non-centred, LKJ prior) and an optional negative price coefficient for every respondent
• part-worths per respondent, importances, RLH and a Sawtooth-style utilities CSV
• a market simulator with credible intervals, price curves and WTP

I compared it with R bayesm on the public camera data. Respondent-level posterior means correlate at 0.997 or more, and the hold-out hit rate is 0.729 vs 0.730. The repo also has a parameter recovery test and lists the targets that are not met yet.

It is pre-alpha. I would like feedback on the model and defaults.