Online IRT — Learning Rate
Control how quickly the model adapts to new student responses.
Learning Rate
Adjust the speed at which the model updates student mastery.
Very Slow
Slow
Moderate
Fast
Very Fast
Minimal changes.
High stability.
High stability.
Slower adaptation.
More stable.
More stable.
Balanced adaptation
and stability.
and stability.
Quicker adaptation.
More responsive.
More responsive.
Rapid changes.
Less stable.
Less stable.
Selected
Moderate
etaFast
1.00
etaSlow
0.40
brake
0.80
η at p=0.5
0.63
Publishing is disabled: read-only until write authorization is settled.
What does this affect?
online-irt has no single learning-rate scalar. Each response moves a learner's mastery by etaSlow + (etaFast − etaSlow) · clamp((brake − p) / brake), so etaFast is the step size for a new learner, etaSlow the floor it decays to, and brake the proficiency where that floor is reached. Faster levels raise both etas and push brake up, so the responsive regime persists further up the scale.