unable to directly obtain probability masses of joint distributions

[Former user]

when i define a multivariate distribution as

joint = lea.Lea.fromValFreqs( ((1,2),10), ((1,3),9), ((2,2),8) ).asJoint('A','B')

i cannot use the pmf or p method since something like

joint.pmf((1,3))

raises an error. instead i currently use

joint.B.given(self.joint.A==1).pmf(3) * A.pmf(1)

should the pmf and p methods be able to handle multivariate arguments or is there another way to directly access the probability masses of multivariate distributions?

[Pierre Denis]

Thank you for reporting! There are two cleaner ways to solve this issue, with the current version of Lea (2.1.2 or before):

1) use Lea cartesian product method to get tuples:

tupleJoint = Lea.cprod(joint.A,joint.B)
tupleJoint.pmf((1,2))  
# 0.37037037037037035

2) (preferred) use Python's named tuples instead of Lea's joint:

from collections import namedtuple
J = namedtuple('J',('A','B'))
joint = Lea.fromValFreqs( (J(1,2),10), (J(1,3),9), (J(2,2),8) )
joint.pmf((1,2))
# 0.37037037037037035

Note that the joint variable works as before (i.e you can access joint.A and joint.B)

In forthcoming version of Lea, I will most probably change the implementation of asJoint method such that it uses Python's named tuples. Then, your expression joint.pmf((1,3)) will work fine.

[Pierre Denis]

Use namedtuple to implement asJoint method, remove Olea class (refs #20)

→ <<cset c45023788d37>>

[Pierre Denis]

Done. asJoint is now implemented with namedtuple (see comment above). Shall be available in Lea 2.2.

[Pierre Denis]

Use namedtuple to implement asJoint method, remove Olea class (refs #20)

→ <<cset 937dec461bd6>>