def test_parse_rvs2():
outcomes = ['00', '11']
pmf = [1 / 2] * 2
d = Distribution(outcomes, pmf)
d.set_rv_names('XY')
with pytest.raises(ditException):
parse_rvs(d, ['X', 'Y', 'Z'])
def test_parse_rvs2():
outcomes = ['00', '11']
pmf = [1/2]*2
d = Distribution(outcomes, pmf)
d.set_rv_names('XY')
with pytest.raises(ditException):
parse_rvs(d, ['X', 'Y', 'Z'])
def brute_marginal_array(d, rvs, rv_mode=None):
"""A brute force computation of the marginal array.
The parameter array tells which elements of the joint pmf must be summed
to yield each element of the marginal distribution specified by `indexes`.
This is more general than AbstractDenseDistribution since it allows the
alphabet to vary with each random variable. It still requires a Cartestian
product sample space, however.
TODO: Expand this to construct arrays for coalescings as well.
"""
from dit.helpers import parse_rvs, RV_MODES
# We need to filter the indexes for duplicates, etc. So that we can be
# sure that when we query the joint outcome, we have the right indexes.
rvs, indexes = parse_rvs(d, rvs, rv_mode, unique=True, sort=True)
marginal = d.marginal(indexes, rv_mode=RV_MODES.INDICES)
shape = (len(marginal._sample_space), len(d._sample_space))
arr = np.zeros(shape, dtype=int)
mindex = marginal._sample_space.index
for i, joint_outcome in enumerate(d._sample_space):
outcome = [joint_outcome[j] for j in indexes]
outcome = marginal._outcome_ctor(outcome)
idx = mindex(outcome)
arr[idx, i] = 1
# Now we need to turn this into a sparse matrix where there are only as
# many columns as there are nonzero elements in each row.
# Apply nonzero, use [1] to get only the columns
nz = np.nonzero(arr)[1]
n_rows = len(marginal._sample_space)
arr = nz.reshape((n_rows, len(nz) / n_rows))
return arr
def brute_marginal_array(d, rvs, rv_mode=None):
"""A brute force computation of the marginal array.
The parameter array tells which elements of the joint pmf must be summed
to yield each element of the marginal distribution specified by `indexes`.
This is more general than AbstractDenseDistribution since it allows the
alphabet to vary with each random variable. It still requires a Cartestian
product sample space, however.
TODO: Expand this to construct arrays for coalescings as well.
"""
from dit.helpers import parse_rvs, RV_MODES
# We need to filter the indexes for duplicates, etc. So that we can be
# sure that when we query the joint outcome, we have the right indexes.
rvs, indexes = parse_rvs(d, rvs, rv_mode, unique=True, sort=True)
marginal = d.marginal(indexes, rv_mode=RV_MODES.INDICES)
shape = (len(marginal._sample_space), len(d._sample_space))
arr = np.zeros(shape, dtype=int)
mindex = marginal._sample_space.index
for i, joint_outcome in enumerate(d._sample_space):
outcome = [joint_outcome[j] for j in indexes]
outcome = marginal._outcome_ctor(outcome)
idx = mindex(outcome)
arr[idx, i] = 1
# Now we need to turn this into a sparse matrix where there are only as
# many columns as there are nonzero elements in each row.
# Apply nonzero, use [1] to get only the columns
nz = np.nonzero(arr)[1]
n_rows = len(marginal._sample_space)
arr = nz.reshape((n_rows, len(nz) / n_rows))
return arr
def test_parse_rvs3():
outcomes = ['00', '11']
pmf = [1 / 2] * 2
d = Distribution(outcomes, pmf)
with pytest.raises(ditException):
parse_rvs(d, [0, 1, 2])
def test_parse_rvs3():
outcomes = ['00', '11']
pmf = [1/2]*2
d = Distribution(outcomes, pmf)
with pytest.raises(ditException):
parse_rvs(d, [0, 1, 2])
