Modify simulating data and the clever randomization

marcdotson / conjoint-ensembles

Using clever randomization and ensembling strategies to accommodate multiple data pathologies in conjoint studies.

MIT License

0 stars 1 forks source link

Modify simulating data and the clever randomization #47

Closed marcdotson closed 3 years ago

marcdotson commented 3 years ago

@jeff-dotson this is yours, working on the simulation-tuning branch.

[x] Allow for pathological heterogeneity across simulated respondents (e.g., not everyone screens on the same attribute levels). This likely requires changes to clever_randomization().

Initial results, for reference:

Model	LOO	Hit Rate	Hit Prob
HMNL	-2732	0.566	0.446
Ensemble	-13.4	0.567	0.402

marcdotson commented 3 years ago

Merged changes from predictive-fit so you should be able to run predictive fit on ensemble output.

jeff-dotson commented 3 years ago

I'm going to hold off on the simulation of covariates. This seems to be a "nice-to-have" feature for the paper. Let's focus on core issues first.

marcdotson commented 3 years ago

Merged changes from predictive-fit so you should be able to run predictive fit on HMNL output as well (plus some modifications to how predictive fit is computed for the ensembles as well).

marcdotson commented 3 years ago

Merged additional changes from predictive-fit so you should be able to again go through the whole pipeline and evaluate fit for you heterogeneity changes, @jeff-dotson.

marcdotson commented 3 years ago

Changes submitted in PR #58.