I feel like there is a problem with indices somewhere. When running the code below where the pairwise potentials are non-informative, the marginals are wrong for a few nodes only (here 1).
For some reason this does not happen all the time though: running the code several times gives different results. I'm using Jupyter Notebook and I relaunch the kernel between each execution.
`import junctiontree.junctiontree as jt
import numpy as np
factors = [
['0'], ['1'], ['2'], ['3'], ['4'], ['5'], ['6'], ['7'], ['8'], ['9'],
['0', '2'], ['0', '3'], ['1', '4'], ['1', '5'], ['1', '6'], ['1', '7'], ['1', '9'], ['2', '5'], ['3', '6'], ['3', '8'], ['4', '6'], ['4', '7'], ['4', '9'], ['5', '6'], ['5', '7'], ['5', '9'], ['6', '8']
]
values = [
np.array([0.51863787, 0.48136213]), np.array([0.37818722, 0.62181278]), np.array([0.54252319, 0.45747681]), np.array([0.61971232, 0.38028768]), np.array([0.6027177, 0.3972823]), np.array([0.39505855, 0.60494145]), np.array([0.43202537, 0.56797463]), np.array([0.58615866, 0.41384134]), np.array([0.51366971, 0.48633029]), np.array([0.51068872, 0.48931128]),
np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]])
]
var_sizes = {'0': 2, '1': 2, '2': 2, '3': 2, '4': 2, '5': 2, '6': 2, '7': 2, '8': 2, '9': 2}
tree = jt.create_junction_tree(factors, var_sizes)
prop_values = tree.propagate(values) #run the junction tree algorithm
marginals = prop_values[:len(var_sizes)] #unitary marginals
marginals = [el / np.sum(el) for el in marginals] #renormalization
for el1, el2 in zip(marginals, values[:len(var_sizes)]):
print(np.round(el1 - el2, 15))`
returns
[0. 0.] [0. 0.] [0. 0.] [0. 0.] [ 0. -0.] [-0.17562959 0.17562959] [0. 0.] [ 0. -0.] [0. 0.] [0. 0.]
I feel like there is a problem with indices somewhere. When running the code below where the pairwise potentials are non-informative, the marginals are wrong for a few nodes only (here 1).
For some reason this does not happen all the time though: running the code several times gives different results. I'm using Jupyter Notebook and I relaunch the kernel between each execution.
`import junctiontree.junctiontree as jt
import numpy as np
factors = [
['0'], ['1'], ['2'], ['3'], ['4'], ['5'], ['6'], ['7'], ['8'], ['9'],
['0', '2'], ['0', '3'], ['1', '4'], ['1', '5'], ['1', '6'], ['1', '7'], ['1', '9'], ['2', '5'], ['3', '6'], ['3', '8'], ['4', '6'], ['4', '7'], ['4', '9'], ['5', '6'], ['5', '7'], ['5', '9'], ['6', '8']
]
values = [
np.array([0.51863787, 0.48136213]), np.array([0.37818722, 0.62181278]), np.array([0.54252319, 0.45747681]), np.array([0.61971232, 0.38028768]), np.array([0.6027177, 0.3972823]), np.array([0.39505855, 0.60494145]), np.array([0.43202537, 0.56797463]), np.array([0.58615866, 0.41384134]), np.array([0.51366971, 0.48633029]), np.array([0.51068872, 0.48931128]),
np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]]), np.array([[0.5, 0.5], [0.5, 0.5]])
]
var_sizes = {'0': 2, '1': 2, '2': 2, '3': 2, '4': 2, '5': 2, '6': 2, '7': 2, '8': 2, '9': 2}
tree = jt.create_junction_tree(factors, var_sizes)
prop_values = tree.propagate(values) #run the junction tree algorithm
marginals = prop_values[:len(var_sizes)] #unitary marginals
marginals = [el / np.sum(el) for el in marginals] #renormalization
for el1, el2 in zip(marginals, values[:len(var_sizes)]):
print(np.round(el1 - el2, 15))`
returns
[0. 0.] [0. 0.] [0. 0.] [0. 0.] [ 0. -0.] [-0.17562959 0.17562959] [0. 0.] [ 0. -0.] [0. 0.] [0. 0.]