If H or Hbc have NaNs, then pymc will raise a confusing error says that sampling couldn't be initialised. This is because every value for y has zero probability, so the logp value is negative infinite.
Adding a check for NaNs before sampling would make it easier to diagnose this problem.
If
H
orHbc
have NaNs, then pymc will raise a confusing error says that sampling couldn't be initialised. This is because every value fory
has zero probability, so the logp value is negative infinite.Adding a check for NaNs before sampling would make it easier to diagnose this problem.