log prices gives divide by zero issues in simulation about pyblp HOT 3 CLOSED

I'm guessing the prices solved for by the simulation have a few zeros. In practice, log terms like this create trouble for numerical estimation routines because a few "extreme" values can create errors. Without more info I can't say much more.

from pyblp.

Actually, I did not even solve for prices. I just initialized the simulation class, using the example in docs, but with log prices instead.

id_data = pyblp.build_id_data(T=50, J=20, F=10)
integration = pyblp.Integration('product', 9)
simulation = pyblp.Simulation(
product_formulations=(
pyblp.Formulation('1 + log(prices) + x'),
pyblp.Formulation('0 + x'),
pyblp.Formulation('0 + x + z')
),
beta=[1, -2, 2],
sigma=1,
gamma=[1, 4],
product_data=id_data,
integration=integration,
seed=0
)

from pyblp.

I see. I think you're using a version of the package (may just be the latest version on pip) where prices and shares are initialized as zero as a placeholder until they can be replaced by Simulation.solve. So this gives the error.

I've overhauled the Simulation class recently (actually in response to #29 that you brought up). This should no longer be a problem in the dev version of the code where prices and shares are initialized as (nonzero) draws from uniform distributions.

Note that my original thoughts about including logs still might be an issue for you.

from pyblp.