Dear all,
I’d like to undestand why I get the dimension mismatch error below.
I’m considering the paradimatic Disaster Model as in the documentation:
disasters_data = array([4, 5, 4, 0, 1, 4, 3, 4, 0, 6, 3, 3, 4, 0, 2, 6,
3, 3, 5, 4, 5, 3, 1, 4, 4, 1, 5, 5, 3, 4, 2, 5,
2, 2, 3, 4, 2, 1, 3, 2, 2, 1, 1, 1, 1, 3, 0, 0,
1, 0, 1, 1, 0, 0, 3, 1, 0, 3, 2, 2, 0, 1, 1, 1,
0, 1, 0, 1, 0, 0, 0, 2, 1, 0, 0, 0, 1, 1, 0, 2,
3, 3, 1, 1, 2, 1, 1, 1, 1, 2, 4, 2, 0, 0, 1, 4,
0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 1])
year = arange(1851, 1962)
with pm.Model() as model:
switchpoint = pm.DiscreteUniform('switchpoint', lower=year.min(), upper=year.max())
early_mean = pm.Exponential('early_mean', lam=1.)
late_mean = pm.Exponential('late_mean', lam=1.)
rate = tt.switch(switchpoint >= year, early_mean, late_mean)
disasters = pm.Poisson('disasters', rate, observed=disasters_data)
Where disasters is a discrete random variable and needs a metropolis sampler:
with model:
step1 = pm.Metropolis(switchpoint)
step2 = pm.NUTS([early_mean, late_mean])
tr = pm.sample(200, tune=100, step=[step1, step2])
So far so good, that is the defaul choice, indeed. However, if I try:
with model:
step1 = pm.Metropolis(switchpoint)
step2 = pm.Metropolis(late_mean)
step3 = pm.NUTS(early_mean)
tr = pm.sample(200, tune=100, step=[step1, step2, step3])
I get a long error, which culminates in:
ValueError: Dimension mismatch; shapes are (2), (1)
Is that something inherent to the NUTS sampler? Or some bug?
Cheers,
M.