Correct choice probability

[Mind-the-Cap]

Adds a divider before using log logic, otherwise the slightly better option will be too heavily chosen

Simple example with the possible divider values :

1 is the current, and we see that the option with utility 14 (vs 18 and 20) is chosen by 98 % of persons, which is not realistic. Given the utilities we use, 10 looks a better option !

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 70.42%. Comparing base (c8c3ac3) to head (504cce6).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #250   +/-   ##
=======================================
  Coverage   70.42%   70.42%           
=======================================
  Files          56       56           
  Lines        2424     2424           
=======================================
  Hits         1707     1707           
  Misses        717      717           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[FlxPo]

I'm not sure about this, on what basis can we conclude that 98 % is not realistic here ?

The problem with the introduction of a global coefficient is that the model is going to give the same results with different utility coefficients : a value of time of 10 €/h and a divider of 1, or a value of time of 100 €/h and a divider of 10, for example. I think this would be confusing for users.

I think the best thing to do would be to divide directly all of the utility of the coefficients. And tell users that they should lower the overall scale of their utility coefficients if they see the model "collapse" on the best option only (which could also be fine in some situations like very car dependant ones).