Inserting random variable into a matrix

1

I’m a beginner with Pymc and unfamiliar with Pytensor. I am trying to write a model with matrices that include both constant and random entries. How can I insert the random entries into a pytensor variable matrix? I find it totally baffling that for a tensorvariable matrix F, F[r,c].set() seems to be a valid function but I can’t find anything about it in the documentation.

2

Hi there!

Sorry you’re feeling baffled. The function you want is pt.set_subtensor, as in:

X = pt.zeros((5, 5))
X = pt.set_subtensor(X[0, 1], 1)
X.eval()
>>> Out: array([[0., 1., 0., 0., 0.],
                [0., 0., 0., 0., 0.],
                [0., 0., 0., 0., 0.],
                [0., 0., 0., 0., 0.],
                [0., 0., 0., 0., 0.]])

You can pass any arbitrary symbolic value to the 2nd argument of set_subtensor, so this works fine:

with pm.Model() as m:
    X = pt.zeros((5, 5))
    a = pm.Normal('a')
    X = pt.set_subtensor(X[0, 1], a)

You can use this to allocate any slice you like, not just single entries. So this also works fine:

with pm.Model(coords={'dim':['a', 'b', 'c', 'd', 'e']}) as m:
    X = pt.zeros((5, 5))
    a = pm.Normal('a', dims=['dim'])
    X = pt.set_subtensor(X[0, :], a)

That replaces the entire first row of X with iid draws from a.

1 Like
3

Thanks for the quick reply!