Vectorization and improving speed for a complex mixture type latent trait model for response times

Modeling
2

So two slightly conflicting suggestions:

  1. Try to avoid declaring variables to store the result of an intermediate computation wherever possible. (Apparently more declared variables can induce substantial slowdown for auto-differentiation)

  2. Try to look for redundant computations and instead of doing them redundantly, store as a declared variable and refer to them where they’re needed.

And presumably you’re aware of the various parallelization options?