Graded Response Model

[mathmonstergo]

Great project! By the way, have you ever considered implementing the Graded Response Model (Polytomous IRT)?

[jplalor]

Thanks! Yes, we've thought about the graded response model but just haven't had the bandwidth to implement it. I'd be very open to reviewing a PR if someone wants to take a pass at it!

[zouharvi]

We do have a continuous scoring model here https://github.com/zouharvi/py-irt that's being used in the subset2evaluate package.

Essentially instead of predicting the binary outcome, it predicts a score in [0, 1]. Maybe that'd suit your needs, @mathmonstergo?

The reason why I didn't file a PR for it just yet is because of some technicalities. Technically the output should be a Beta distribution but for us the Normal distribution worked better for which we don't have a good explanation. The way it works is that instead of predicting the Bernoulli p and compare with 0 or 1 coming from the observations, we predict the Normal distribution mean and compare it with the observed score.