Metropolis sampler do not sample (traceplot gives flat plots)

ok, the output of your suggestion is:

Intercept 0.0
age -7.82669381218681
ages -7.82669381218681
hours -7.82669381218681
educ -7.82669381218681
y -15638.093540616184

all the same.

But more: now not even the copy-paste of the code in here works (I got the error about the bad initial energy).

I tried to delete the Theano cache, but nothing changes.

Is there any kind of cache of PyMC that I can flush?