How to increase weights of conditional log probability

Questions v5
8

It will not override anything, it’s just added to the total model logp.

Here is a simple example and some code so you can investigate if things are doing what you want them to:

import numpy as np
import pymc as pm
from pymc.model.fgraph import clone_model

with pm.Model( ) as m1:
  x = pm.Exponential("x", 1)
  y = pm.Poisson("y", x, observed=[0, 1, 2])

with clone_model(m1) as m2:
  # Just add one to whatever the model joint-probability is
  pm.Potential("potential", pm.math.ones((1,)))

print(m1.point_logps())  # {'x': -1.0, 'y': -3.69}
print(m2.point_logps())  # {'x': -1.0, 'y': -3.69, 'potential': 1.0}
np.testing.assert_allclose(
    m1.compile_logp()({"x_log__": 0}) + 1,
    m2.compile_logp()({"x_log__": 0}),
)
1 Like