Fix missing args in tfe.stats.RunningVariance.from_shape

[chrism0dwk]

Both the event_ndims and name arguments of the class method tfp.experimental.stats.RunningVariance.from_shape were missing, causing TypeError: from_example() got an unexpected keyword argument 'event_ndims'.

[chrism0dwk]

N.B. this fix sorts out the same TypeError when using RunningVariance.from_example().

[SiegeLordEx]

How did you get the original error? RunningVariance is expected to have event_ndims=0, because it treats each element as an independent (co)variance estimation problem.

[chrism0dwk]

@SiegeLordEx The original problem arose with

tfp.experimental.stats.RunningVariance.from_example(np.array([1., 1., 1., 1.]))

The fundamental issue is with the Python MRO with RunningVariance derived from RunningCovariance:

1. RunningVariance.from_example calls RunningCovariance.from_example;
2. RunningCovariance.from_example calls RunningVariance.from_shape;

The issue is caused because inside RunningCovariance.from_example, cls is a reference to the derived class, not (as I think the author assumed) the base class. Thus RunningCovariance.from_example assumes that RunningVariance.from_shape takes the same arguments as its own RunningCovariance.from_shape method.

Took me a while to revise my Python MRO semantics, but here's the MRE:

class Foo:
    @classmethod
    def bar(cls, baz, quz):
        print("Foo.bar cls:", cls)
        cls.quux(baz, quz)

    @classmethod
    def quux(cls, corge, grault):
        print("Foo.quux cls:", cls)

class Garply(Foo):
    @classmethod
    def bar(cls, baz):
        print("Garply.bar cls:", cls)
        super().bar(baz, quz)

    @classmethod
    def quux(cls, corge, grault):
        print("Garply.quux cls:", cls)
        super().quux(corge, grault)

If Garply.bar's signature is incompatible with Foo.bar's signature (e.g. missing an argument), then the call stack fails.